70764 MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Sustainable Development Sector Management Unit Latin America and the Caribbean Region Document of the World Bank ABBREVIATIONS AND ACRONYMS AISP �reas Integradas de Segurança Pública (Integrated Areas of Public Security) BOPE Batalhão de Operações Policiais Especiais (Special Police Op- erations Batallion – Rio de Janeiro Elite Squad) CAGED Cadastro Geral de Empregados e Desempregados (General Registry for Employed and Unemployed Persons) CCT Conditional Cash Transfer CISP Circunscrições Integradas de Segurança Pública (Integrated Circuits of Public Security) COMPSTAT Computerized Statistics (New York Police Department Sys- tem) DATASUS Base de Dados do Sistema Único de Saúde (Brazilian Health System Database) GDP Gross Domestic Product GEPAR Grupo Especial de Policiamento em �reas de Risco (Special Police Group for High-Risk Areas) IBGE Instituto Brasileiro de Geogra�a e Estatística (Brazilian Insti- tute for Geography and Statistics) IGESP Integração e Gestão da Segurança Pública (Integration and Management of Public Security) INESC Instituto para Estudos Socioeconômicos (Institute for Socio- economic Studies) Infocrim Sistema de Informação Criminal (Crime Information System) ISPCV Instituto São Paulo Contra a Violência (São Paulo Institute Against Violence) MIC Middle Income Countries OECD Organization for Economic Co-operation and Development OLS Ordinary List Squares PAHO Pan American Health Organization PRONASCI Programa Nacional de Segurança Pública com Cidanania (Na- tional Public Security Program) RAIS Relação Anual de Informações Sociais (National Report on Social Information) RISP Regiões Integradas de Segurança Pública (Integrated regions of Public Security) RMSP Região Metropolitana de São Paulo (São Paulo Metropolitan Region) SEDS Secretaria Estadual de Defesa Social (Secretariat of Social Defense) SIM Sistema de Informação sobre Mortalidade (Mortality Informa- tion System of the Ministry of Health) UPP Unidade de Polícia Paci�cadora (Police Paci�cation Unit) WHO World Health Organization Report Vice President: Hasan A. Tuluy Country Director: Deborah Wetzel Sector Director: Ede Jorge Ijjasz-Vasquez Sector Manager: Maninder Gill Sector Leader: Gregor Wolf/Paul Kriss Task Team Leader: Rodrigo Serrano-Berthet Co-Task Team Leader: Laura Chioda TABLE OF CONTENTS ACKNOWLEDGEMENTS 9 OVERVIEW 11 Chapter 1. Trends and Shifts in Violent Crime in Brazil 17 Gradual decline at the national level 18 Marked Di erences Across Regions 19 Persistence of youth violence 22 Chapter 2. Reducing Crime in Brazil: The Factors Driving the Change and Regional Heterogeneity 27 Potential Drivers of Change 28 Channels through which these Factors a ect the homicide rate 30 Factors that correlate with national trends 32 Factors that correlate with changes across and within regions 39 Analysis of the “Residuals�—a Potential Role for Policy? 45 Chapter 3. The Success of the Southeastern States: A Role for Policy? 51 Reforming the Police and the Public Security System 54 Reducing Environmental Risks: Guns, Alcohol, and Unsafe Urban Spaces 58 Targeting Hotspots and Populations at Risk through Multisectoral Strategies 63 Integrated and Multi-Stakeholder Subnational Citizen Security Strategies 69 Applicability to other states? 73 Chapter 4. Reducing Crime among Youth: The Role of Education Policy 77 Channels 79 Evidence from São Paulo 81 Summary of results 85 Conclusions and way forward 77 REFERENCES 95 ANNEX 105 Annex I. Data Sources 106 Annex II. Kernel density distributions tests for equality 108 Annex III. Fixed e ects regressions of the Log Homicide rate per 100,000 for Brazil and its regions 110 Annex IV. Oaxaca Blinder Decomposition 2008 v. 2003 111 Annex V. Changes in Homicide rates per 100,000 inhabitants from 2003 to 2008 by state and state capital 112 List of tables Table 1. Homicide Rates and Black Victimization Index by Region 23 Table 2: Elasticities of macro, contextual and micro factors and the homicide rate in Brazil 1998-2008 32 Table 3: Correlation Coe cients and changes in risk factors and homicide rate across regions (1998-2008) 39 Table 4: Changes in macro, context and micro factors by region 2003 v.2008 42 Table 5: Frequency-Weighted Largest Positive 100 Overall Residuals 46 Table 6: Frequency-Weighted Largest Negative 100 Overall Residuals 47 Table 7: Residuals ranking (negative to positive) 48 List of graphs Graph 1. Brazil Homicide Rates 1980–2008 19 Graph 2. Brazil Kernel Densities of Homicides 2003–2008 19 Graph 3: Homicide Rate Trends in Brazil and Latin America 19 Graph 4. Homicide Rates by Region, 1997–2008 20 Graph 5. Kernel Densities of Homicides by Region 21 Graph 6. Male Homicides Rates in Brazil per Race and Age, 2006 24 Graph 7: Decomposition of Changes Homicide rates for Brazil and its regions 34 Graph 8: Decomposition of Changes of Endowments in Factors 36 Graph 9: Decomposition of Changes of Characteristics in Factors 37 Graph 10. Contributions of endowments for Brazil, the Southeast region, and the Northeast region 2008 v.2003 38 Graph 11: Contribution of coe cients for Brazil, the Southeast region, and the Northeast region 2008 v.2003 43 Graph 12: Contribution of coe cients for Brazil, the Southeast region, and the Northeast region 2008 v.2003 44 Graph 13. Homicide Rates for State and State Capital of São Paulo, Rio de Janeiro, and Minas Gerais, 1998-2008 52 Boxes Box 1. IGESP Impact Assessment Provides Evidence of Strong E ect on Crime 56 Box 2. The Impact of Guns on Youth 58 Box 3. Neighborhood Disorder and Crime Victimization in São Paulo 62 Box 4. Impact Evaluation of Fica Vivo 66 Box 5. Extending the Right to live in Peace to Favela Residents by Changing the Rules of the Game for Rio de Janeiro’s Drug Tra ckers: Unidades de Polícia Paci�cadora (UPP) 67 Box 6. Municipal-level strategies for crime prevention in Diadema 69 Box 7. Exploring the Perceptions of Frontline Workers about the Drivers of Crime Reduction – A Positive Deviant Analysis in Belo Horizonte 72 Box 8. Policy Details (1) Conditional Cash Transfers 82 Box 9. Policy Details (2): Change in School Shifts 84 ACKNOWLEDGEMENTS This study was led by Rodrigo Serrano-Berthet and Laura Chioda. Other members of the team include João Pedro Azevedo, Valentina Calderón, Flávia Carbonari and Jim Shyne. Back- ground papers were prepared by Rodrigo Soares, João Mello, Fundação João Pinheiro, Le- andro Piquet, and Erik Alda. Melissa Zumaeta, Priscilla Burity, Christian Borja, Jessica Varat and Darwin Marcelo provided technical inputs. Maninder Gill provided overall guidance and coordination. Makhtar Diopp, Sameh Wahba and Tito Cordella contributed with valuable and detailed feedback throughout the process. The team is also grateful to the reviewers Andrew Morrison, William Sedlacek and Alexandre Marc, and to Maribel Cherres for the excellent administrative support. OVERVIEW MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Brazil is a country of strong internal contrasts; crime and violence is no exception. Over the course of this decade, the southeast region has experienced a massive decline in its homicide rate, which combined with the rapid increase in the northeast and north regions have led to an overall modest decline in the national average. In other words, while some Brazilians have become substantially safer others have seen their levels of safety deteriorate considerably. Although modest, the recent decline in the national average represents an important change in trend. The annual homicide rate, the leading indicator for levels of crime and violence, has gradually come down from its peak of 28.9 in 2003 to 25.2 in 2007,26.4 in 2008, and has had a slight increase more recently, to 26.8 in 2009 and 27.2 in 2010 (per 100,000 inhabitants). Despite the marginal increase of the past two years for which data is available, the overall trend over the decade is still a signi�cant accomplishment. Several decades of continued increases in violence have made Brazil an outlier with respect to similar middle-income countries inside and outside the region. While Brazil’s homicide rate has always scored much higher than the Latin American average, 2008 was the �rst time in which it scored lower, due in part to the growing problem of violence in other countries in the region. This shift in the trajectory of crime over the last decade signals that important forces are working to drive homicides down. Still, the overall national decline masks an enormous heterogeneity in trends across regions, states and municipalities. Unlike other countries, such as Colombia or the United States, which have experienced similar drops in crime, crime trends at the subnational level in Brazil have not accompanied the national decline in a homogeneous manner. In fact, the dramatic decline in homicides in the Southeastern region contrasts with the equally dramatic increase in the Northern and Northeastern regions. While states such as São Paulo saw their homi- cide rate drop by 67% percent between 2000 and 2010, others like Bahia (+303%), Alagoas (+160%), and Pará (+252.9%) witnessed their homicide rates double or triple during the same period. Pernambuco is an exception to these regional trends (-28%). A particularly worri- some trend is the persistence, and deterioration in some respects, of violence among male youth. Risk of involvement in homicide, as victim or perpetrator, is strongly associated with youth, male gender and nonwhite race. Afro-Brazilians are victims of homicide to a dispro- portionate degree: they are twice as likely to be murdered as their white peers. What factors were driving the overall crime decline in Brazil? Why is violent crime declining in some states while it is increasing in others? What types of interventions could help to reduce youth violence? These are the questions that motivated this report. Understanding what has gone right to bring crime down in some areas during the past decade is crucial to tackling the challenges presented by the new decade, including the more recent marginal increase in the national homicide rate. The purpose of this report is to enhance that understanding. To do so, we examine the determinants of the crime shift at the national level, review the experience of the high-performing states, and generate new evidence on the impact of education policies on youth violence prevention. OVERVIEW As a �rst step toward understanding the drivers behind the change in homicide rates, we es- timated the correlation between six well-known risk factors and homicide rates. We compiled a unique database with 32,000 observations (in contrast to other studies in Brazil that have used at most 300 observations). We found that the decline at the national level is strongly correlated with the reduction of the male youth cohort (ages 15 to 19), the reduction in in- equality, and to a lesser extent the reduction in the dropout rate among public high school students. There is much variation across regions and states. The Southeastern region, particularly São Paulo, has been the driver of the national drop in violent crime between 2003 and 2008. If the States of São Paulo, Minas Gerais and Rio de Janeiro are removed from the national average, the homicide rate would have increased by 29 percent instead of having declined by 7 percent over that period. This is due mostly to worsening trends in the Northeastern (NE) and Northern regions, where rates climbed from 18 to 28.8 per 100,000 (NE) and from 22.50 to 32.97 (North) between 2003 and 2008. The experience of the Southeast o ers important lessons for states with deteriorating con- ditions. First, there is no silver bullet to bring down violent crime. Similar to �ndings in the literature on the North American crime drop during the 1990s, in Brazil no single factor is su cient to explain the overall decline. While factors that are not directly related to subna- tional policies appear to have played an important role (i.e., demographic change and re- duced income inequality), a wide range of policies also appear to have contributed to this decline. The report reviews some of the key policies implemented in these states and pres- ents evidence about the results achieved. What are the policies that have distinguished the Southeast from the rest? At the top of the list are results-oriented policing, gun and alcohol control, and programs targeting youth at risk and hot-spots of crime and violence, all guided by integrated and multi-stakeholder citi- zen security subnational strategies. These policies are not o ered as prescriptions of what governments should do since the evidence basis on their e ectiveness is not strong enough to establish causal relationships; even when they are, their external validity or transferability is not established. Instead they are o ered as examples of policies that are worth learning more about through more robust analytical work, impact evaluations and other forms of controlled policy learning. In particular, this learning agenda should include �ve themes that appear to have been important in the policy experience of the southeastern states: š Reduce youth fragility by addressing risk factors and strengthening protective fac- tors. In particular, the study on correlates suggests that keeping youth in school could be an important protective factor. This report presents original �ndings on the impact of conditional cash transfers (CCTs) for schooling on violent crime, focusing on the marginal enrollment of 16- to 17-year-old males attributable to the CCTs. It �nds that each additional student receiving a CCT at a given state school is causally related MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME to 0.39, 0.46, 0.25, 1.50, and 0.78 percent declines in all crimes, robberies, violent crimes, drug-related o enses, and crimes against minors, respectively. š Develop comprehensive crime prevention strategies that combine territorially based interventions targeting crime and violence hotspots with generalized policies that reduce risks across the board. š Sequence and integrate crime control and crime prevention activities in hotspots of violence. On the one hand, a paci�ed territory facilitates social development activities oriented to address crime and violence risk factors. On the other hand, manifest so- cial development gains in the immediate wake of public security improvements can play a crucial role in making the peace process socially sustainable. š Introduce a results orientation that rewards innovation and performance account- ability, as through results oriented policing. š Build monitoring and evaluation systems that enable shared diagnostics and poli- cy learning and experimentation among the key actors involved in the public safety agenda. The future research agenda should also focus on understanding the dynamics of violent crime in the states that have su ered rapid increases in violent crime, particularly in the Northeast and North regions. From a policy perspective, looking ahead, Brazil needs to con- solidate and deepen the gains made in the Southeast and tackle head-on the challenge of the Northeast and North regions. Fortunately, there is good news coming from the North- east as well. Pernambuco and its capital city, Recife, which for much of the past decade had the dubious distinction of being by far the country’s most violent major metropolitan area, have seen homicide rates decline since 2008. Perhaps not coincidentally, Pernambuco has implemented substantial public security reforms based on successful experiences from the Southeast. Other Northeastern states, including Bahia, have pledged to follow suit. The Federal Government has launched a renovated National Public Security Program (Pro- grama Nacional de Segurança Pública com Cidadania, PRONASCI) and is investing in infor- mation systems to better understand data on crime and public security and develop stronger impact evaluations of potentially transformative programs such as the UPP and UPP Social in Rio de Janeiro. A new legislation (SINESP) has been approved with the objective of pres- suring states to integrate all police, health and justice sectors databases, systematizing the methodologies on data collection used across the country, and promoting the use of evidence in the design of citizen security policies. Roadmap of the report. The report is organized in four chapters. Chapter 1 sets the stage for the issues covered in the report. Chapter 2 estimates the correlation of the change in crime in Brazil and across regions and states. Chapter 3 reviews the evidence on the policies implemented to reduce crime and violence in São Paulo, Minas Gerais and Rio de Janeiro. OVERVIEW Chapter 4 presents �ndings on the impact of school enrollment on youth crime and violence prevention. The last section concludes and summarizes key lessons. Crime and violence are two distinct concepts. While highly correlated, “crime� and “violence� may result from di erent causal channels that predispose individuals to commit crimes or acts of violence in di erent measure. Sedlacek (2010) de�nes crime as the means to obtain a goal or gain, while violence in many cases is an end in itself. Committing a crime can be mod- eled as a rational decision, taken after the valuation of its costs and bene�ts. Committing an act of violence, on the other hand, can result both from rational decisions and from exposure to risk factors that incline individuals toward this type of behavior. In this report we use “homicide� and “violent crime� interchangeably. It should be noted, however, that violent crime is a broader concept that includes all criminal homicides as well as any crime in which the o ender uses or threatens to use violent force upon the victim. The criminal threat or use of violent force covers assaults, rapes, robberies, attempted ho- micides, kidnappings, and other crimes. In this report we concentrate on homicide rates for three main reasons. First, homicides serve as rough proxies for all incidents of criminal vio- lence better than any other single crime statistic. (Fajnzylber, Lederman and Loayza, 2002; Kovandzic and Vieraitis, 2006). Second, homicide rates are measured more completely and more accurately than any other crime indicator, and thus su er less from measurement er- ror problems. Third, homicide data is available for all Brazilian municipalities for the period 1998-2008, which allows us to construct a municipal level panel to calculate elasticities be- tween demographic and socioeconomic factors and the homicide rate. CHAPTER 1. TRENDS AND SHIFTS IN VIOLENT CRIME IN BRAZIL MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME This chapter describes three trends that lead logically to the three research questions at the core of this report: What is driving the reduction in violent crime in Brazil? Why is there so much variation across regions? What can we do to reduce youth violence? GRADUAL DECLINE AT THE NATIONAL LEVEL Starting in the mid-1980s and for two decades, Brazil saw its homicide rate grow to lev- els atypical of any middle-income country. Brazil’s homicide rate grew at a much faster annual rate than any other middle-income country not a ected by war, except for South Africa where violence increased dramatically after 1994. From the mid-1980s until 2003, the homicide rate grew an average of 20 percent per year. The Brazilian literature has suggested a number of factors for this increase. Among the most salient ones are increased urbaniza- tion, volatile and unstable economic conditions that led to high levels of income inequality and social exclusion, an upward shift in population demographics, and rapid expansion of drug-tra cking groups in the main urban areas and the associated proliferation of �rearms (Landman 2003; Briceño-León and Zubillaga 2006). The development and welfare costs for Brazilians have been enormous. Homicide has been the leading cause of death among youth (between the ages of 15 and 24) since the 1980s, as well as the main cause of lost life-years in the country. Direct costs of violent crime are estimated at about �ve percent of the country’s gross domestic product (GDP), calculated as the sum of public and private investments directed toward security rather than other areas; lost business investments; and changes in patterns of consumption due to security concerns (Cerqueira et al. 2007). As a result, more than half of Brazilians recently reported feeling unsafe in the cities and neighborhoods where they live (IBGE 2010). Since 2003, Brazil’s homicide rates have �nally begun to drop. The annual homicide rate, the leading indicator for levels of crime and violence, has come down from its peak of 28.9 per 100,000 population in 2003 to 26.4 in 2008 (see Graph 1). Graph 2 presents kernel den- sity distributions of log homicide rates per 100,000 from all 5,565 municipalities in Brazil in 2003 (red line) and 2008 (broken blue line). The shift down and toward the origin represents a global fall in the means of these rates from the earlier to the later time period. A Kolmogorov- Smirnov test reveals the shift to be statistically signi�cant. CHAPTER 1. TRENDS AND SHIFTS IN VIOLENT CRIME IN BRAZIL Graph 1. Brazil Homicide Rates 1980–2008 Graph 2. Brazil Kernel Densities of Homicides 2003–2008 30 Brazil 6 25 4 Homicide Rate Density 20 2 15 0 0 2 4 6 10 1980 1990 2000 2010 Log Homicide Rate per 100000 Year 2003 2008 Source: Datasus Source: Datasus Brazil is improving, but it still has a long way to go. Brazil is doing better than the rest of Latin America. In 2008, for the �rst time in decades, the average homicide rate in Brazil was lower than the average for Latin America and the Caribbean (see Graph 3). However, when compared to countries at similar levels of economic development, Brazil continues to display signi�cantly higher levels of violence. Organization for Economic Co-operation and Develop- ment (OECD) countries (1.62, in 2008), such as India (2.87), Russia (14.2), Argentina (5.8) and even Mexico (12) all had much lower homicide rates over the past decade. Graph 3: Homicide Rate Trends in Brazil and Latin America, 1990-2010 30 25 Homicide Rate per 100,000 people 20 15 10 5 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Year Homicide Rate Brazil Homicide Rate LAC Source: DATASUS (BR) National sources (LCR) MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Many factors could be driving this reduction in violent crime. In fact, the few studies that have tried to explain instances of crime decline at the national level (such as the crime shift in the US or Canada) have made strong arguments in favor of multicausality (Blumstein and Wallman 2006; Zimring 2007). Zimring concludes his review of the crime decline in the US with the statement that “the crime decline of the 1990s was a classic example of multiple causation, with none of the many contributing causes playing a dominant role�. As a �rst step toward better understanding the case of Brazil, Chapter 2 estimates the correlation between six well-known risk factors and changes in homicide rates during 1997–2008. MARKED DIFFERENCES ACROSS REGIONS Depending on the region where Brazilians live, they have seen their environment become considerably safer or alarmingly more violent over the last ten years. In other words, there is strong spatial heterogeneity in the distribution and evolution of violent crime in Bra- zil. Speci�cally, the Southeast is moving in a sharply di erent direction than the Northern and Northeastern regions. Indeed, only in Southeastern Brazil are homicide rates falling, while across the rest of the country, trends are deteriorating more or less precipitously. Graph 4. Homicide Rates by Region, 1997–2010 Brazil by Region 40 Homicide Rate per 100000 30 20 10 1996 2000 2005 2010 Year Southeast North Central-West Northeast South Source: Datasus CHAPTER 1. TRENDS AND SHIFTS IN VIOLENT CRIME IN BRAZIL All regions in Brazil, except the Southeast, experienced increases in the volume and rate of homicides between 2003 and 2008. The nationwide kernel density distribution for crime in Brazil presented at the beginning of this chapter, comparing 2003 to 2008 mean log homi- cide rates for all of the country’s municipalities, showed a decline not only among average homicide rates from one period to the next, but also a shift to the left in their overall distribu- tion, with more than half of the country’s municipalities in 2008 falling below the 2003 mean. The same analysis disaggregated by region tells a very di erent story, bringing attention to the great regional heterogeneity underlying the decline in homicide rates at the national level. Graph 5 shows that falling homicide rates concentrated in Southeastern Brazil have been driving the national trend. Graph 5. Kernel Densities of Homicides by Region North East North South East 1.5 1.5 1.5 1 1 1 Density Density 0.5 0.5 Density 0.5 0 0 0 0 2 4 6 0 2 4 6 0 2 4 6 Log Homicide Rate per 100000 Log Homicide Rate per 100000 Log Homicide Rate per 100000 2003 2010 2003 2010 2003 2010 South Central-West 1.5 1.5 1 1 Density Density 0.5 0.5 0 0 0 2 4 6 0 2 4 6 Log Homicide Rate per 100000 Log Homicide Rate per 100000 2003 2010 2003 2010 Note: Bandwidth 0.25. Source: Own estimations based on DATASUS/SIM data and IBGE. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME The Southeast experienced by far the sharpest decline in homicides over the 2003–2008 period, dropping from an average of 36.3 per 100,000 in 2003 to 21.74 per 100,000 in 2008. In fact, this decline was not homogeneously distributed even among states within the Southeastern region. On the one hand, São Paulo and Rio de Janeiro experienced spectacu- lar declines in homicide rates starting in late 2002. In São Paulo, the rate fell from 36.03 per 100,000 in 2003 to roughly 14.96 per 100,000 in 2008, with this declining trend begin- ning in 1999. In Rio de Janeiro, homicide rates fell from 52.69 to 33.98 per 100,000 over the same period, while rates in Minas Gerais declined only moderately from 20.97 to 19.84 per 100,000. Homicide rates in Espírito Santo actually increased over this period, from 50.45 per 100,000 in 2003 to 56.40 per 100,000 in 2008. The kernel density graphs by region also show that the Northeastern and Northern regions experienced an increase in crime rates and the incidence of higher crime rates, as depicted by heavier left-side tails, while in the South and Center-West homicide rates remained largely steady between 2003 and 2008. In the Northeast the homicide rates between 2003 and 2008 rose from 18.02 to 28.78 per 100,000, while in the North they climbed from 23.5 per 100,000 in 2003 to 33.5 per 100,000 in 2008. From 2003 to 2008 alone, heterogeneity in the percent change in homicide rates across states was substantial: São Paulo showed the largest rate reduction (a staggering 59 per- cent), while Amazonas, Maranhão, Paraíba, Rio Grande do Norte, Bahia, Alagoas and Pará experienced increases of more than 50 percent in their homicide rates over this �ve-year period. Annex II shows the di erences in homicide rates by state from 2003 to 2008. The decline in the state capitals of the Southeast was even more striking than that at the state level. The City of São Paulo reduced its homicide rate by 76 percent (2000–2008), Belo Horizonte by 32.5 percent (2003-2008), and Rio de Janeiro City by about 25 percent (2003–2010). The wide variation across regions and states can be explained largely by di erences in eco- nomic performance, demographics, degrees of urbanization, and school attainment among at-risk youth. Chapter 2 analyzes the factors that help to explain the observed heterogeneity across regions and states. PERSISTENCE OF YOUTH VIOLENCE The most common face of the victims of violence in Brazil is that of a poor, young male Afro-Brazilian. This is the case now and has been for a long time. Ninety-two percent of homicides have men as their victims. Young men and adolescents in the 15 to 24 age cohort represent 18.6 percent of the population and 36.6 percent of all homicide victims; 15- to 29–year-old males represent 25.5 percent of the population and 54.7 percent of homicide victims (2007 PAHO data). Afro-Brazilian youth are more than twice as likely to be victims of CHAPTER 1. TRENDS AND SHIFTS IN VIOLENT CRIME IN BRAZIL homicide than are white Brazilians of the same age. For example, data for 2000 show that the homicide rate for young Afro-Brazilians was more than 74 homicides per 100,000, whereas the rate for young white Brazilians was 41 homicides per 100,000 (Silva and Begin 2005). Moreover, poor Afro-Brazilian youth are disproportionately a ected by the levels of violence (INESC 2007). Despite a slight decline in the homicide rate among youth since 2003, the percentage of total homicides accounted by youth has slightly increased. While in 2003 a youth was 2.52 times more likely to be victimized than the rest of the population, in 2008 the likelihood increased to 2.58 times (Sangari 2011). This is despite the fact that homicide rates among youth have declined slightly. The homicide epidemic among Afro-Brazilians continues to worsen. Brazilians of African descent have historically been over-represented among homicide victims, and this trend shows no sign of abating nationwide. Only in the far Southern region, where a relatively small percentage of the nation’s Afro-descended population resides, have rates of black victimiza- tion begun to fall, while in major Afro-Brazilian population centers such as the Northeast, rates have skyrocketed in recent years, from 185.7 per 100,000 in 2002 to 375.2 per 100,000 in 2008 (see Table 1). Table 1. Homicide Rates and Black Victimization Index by Region Homicide Rates per 100,000 Black Victimization Index Region White Black 2002 2005 2008 2002 2005 2008 2002 2005 2008 North 17.8 13.6 12.9 32.1 27.9 36.1 79.9 80.2 169.6 Northeast 8.2 7.7 9 23.4 26.3 37.8 185.7 227.3 375.2 Southeast 26 19.2 14.6 50.5 37.7 28.6 94.3 71.1 73 South 17.7 19.9 23.9 18.7 22.8 22.6 5.7 16.1 -6.4 Center-West 20.6 17.2 17.4 33.7 34.8 39.3 63.3 85.7 127.8 Brazil 20.6 17.1 15.9 30 31 33.6 45.8 67.1 103.4 Source: Walselfisz 2011. Mapa da Violência 2011 – Os Jovens do Brasil. The following graph shows that Brazilian men of color (“Negra� below) su er consistently higher homicide rates than their European-descended peers across the lifecycle, with Afro- Brazilian youth in the 12 to 35 age cohort more than twice as likely to be victims of homicide than white Brazilians of the same age. As a response to this concerning and historically neglected trend, at the end of 2012 the federal government launched a National Plan for the Prevention of Violence Against Black Youth, also called Youth Alive (“Juventude Viva�). MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Graph 6. Male Homicides Rates in Brazil per Race and Age, 2006 300 250 200 150 100 50 0 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70 72 74 76 78 80 e mais Branca Negra Source: INESC 2010. There is an urgent need to build evidence on what works to reduce youth violence in Bra- zil. There is a growing evidence base on what works in developed countries. However, the lit- erature in Brazil is very incipient. Chapter 4 reviews the international literature and explores the contribution that education policy can have on youth crime and violence. CHAPTER 1. TRENDS AND SHIFTS IN VIOLENT CRIME IN BRAZIL CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME This chapter presents evidence on some of the factors correlated with changes in violent crime between 1998 and 2008. The chapter estimates the correlation between six risk fac- tors and national homicide trends at the national, regional and state levels, using a novel data set for all Brazilian municipalities. The results presented here are part of background paper on the trends of crime rates across regions in Brazil. POTENTIAL DRIVERS OF CHANGE Violent crime is an exceedingly complex phenomenon, caused by multiple factors, which has given birth to a large body of literature in economics, sociology, and psychology, among other disciplines. An area of agreement among the di erent disciplines is that there is no one direct cause in personal development that leads automatically to violent behav- ior. Instead, there are characteristics of an individual’s biology, personality, and environment that impose stresses, which increase the risk that he or she will perpetrate or experience violence. The accumulation of these stresses, or risk factors, is associated with an increased tendency to be a victim or a perpetrator of violence. Conversely, protective factors can be understood as characteristics of an individual and his/her environment that strengthen the capacity to confront stresses without the use of violence. Studies in the US and elsewhere suggest that the accumulation of risk and protective factors is more influential than the im- pact of any one risk factor (see citations in World Bank 2010).1 We focus on six socio-economic factors that can be classi�ed into 3 di erent categories: Macro Factors t Economic cycles (business cycle measured by the GDP) t Income inequality (measured by the GINI coefficient) t Labor market conditions (formal jobs creation and destruction) Context Factors t Urbanization t Population density Micro Factors t Demographic factors include e.g. age and sex (in particulat the cohort size of young males) t Education as protective factor, by considering the role of high school dropout rates and their correlation with crime. 1 The most common conceptual model to organize preventative and risk factors has been the ecological risk model. This model identi�es four levels though which speci�c factors can influence whether or not individu- als engage in criminal behavior: societal, community, relationship/interpersonal and individual (WHO 2002). While this is a compelling model, we do not use it to frame our analysis because we do not have individual level data but rather municipal aggregates. CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY The suggested hierarchical grouping of factors provides a simple and practical organiza- tional structure for discussing the results of this chapter. Due in part to the abundance and variety of municipal level data available for the entire country, the set of factors considered is not an exhaustive list of the main determinants of crime. However it does allow us to mea- sure crime elasticities with respect to some variables that are or can be influenced by policy, and have received considerable attention in the policy debate as well as in the literature. Macro factors include economic conditions (business cycle measured by the GDP), income inequality (measured by the Gini coe cient), and labor market conditions (formal jobs cre- ation and destruction). Contextual factors include measures of urbanization and population density. Micro factors include demographic characteristics such as age and sex (in particular the cohort size of young males); in addition we capture the role of education as a protective factor, by considering the role of high school dropout rates and their correlation with crime. In discussing some of the results we will also comment on inertia of crime rates. This decomposition allows estimating the contribution of endowments and coe cients to the crime reduction observed between 2003 and 2008. The analysis identi�es three of these factors as powerful predictors of the reduction of homicide rates in this time frame at the national level, namely, changes in the distribution of income, measured here by the Gini coef- �cient; the demographic reduction observed in the cohorts of young males aged 15-19, and the reduction in school dropout rates. These three changes emerge as the most important factors associated with the decline in the homicide rate between 2003 and 2008 for Brazil as a whole. At the regional level we observe wide heterogeneity in how these factors a ect crime rates. The correlations of all of these factors with respect to the homicide rate at the re- gional level vary substantially in sign and magnitude. The Southeast and Northeast appear to have divergent trends in crime rates following 2003. The Oaxaca decomposition of these two regions suggests that the potential key factors driving the di erences are changes in the coe cients of urbanization, the concentration of income measured by the Gini coe cient and the high school dropout rate. Most notably, changes in coe cients for urbanization are particularly important for the Northeast. We carry out a detailed residual analysis that allows us to make inferences on how the fac- tors not controlled for in the estimations explicitly correlate between and within the regional heterogeneity of homicide rates in Brazil for the 1998–2008 period. We present weak and only suggestive evidence on the unexplained component and how it relates to variables that are in principle controllable by policy. These variables a ect crime either by raising the ex- pected cost of crime or by deterring criminals. The rest of the chapter is organized as follows. In the second section we present the chan- nels through which these factors a ect the homicide rate. In the third section the panel regression analysis results and the elasticities of these factors with respect to the homicide rate are discussed, followed by the Oaxaca-Blinder decomposition results. In the fourth sec- MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME tion we present the regional elasticities and decomposition results. In the �fth and �nal sec- tion we present the analysis of residuals. CHANNELS THROUGH WHICH THESE FACTORS AFFECT THE HOMICIDE RATE The relationship between these di erent factors and crime can be either positive or nega- tive. In some cases, opposing forces are involved when these factors increase or decrease. We present a description of the mechanism through which these factors can a ect crime rates, and conduct empirical analyses to better understand the magnitude and signs of the correlations. Macro factors Economic conditions (business cycle measured by the GDP). There are two opposing forces that operate when the relationship between the business cycle and crime is estimated. On the one hand, crime could be countercyclical, because it is a substitute for legitimate sources of income (Bushway, Cook and Phillips 2010). But crime can also be pro-cyclical if one as- sumes that more goods and services produced in the economy represent more opportuni- ties for criminal activity. The �rst assumption is likely to outweigh the second one for crimes involving direct �nancial motivation such as burglary, robbery and auto theft, but is less important for homicide, assault and rape (Levitt 2004). Income inequality (measured by the GINI coe cient). The literature has found a strong as- sociation between levels of inequality and crime rates. High levels of inequality place poor individuals who have low returns from market activities in close proximity to high-income in- dividuals who have goods worth taking (Kelly 2000). The e ects of inequality on crime appear to be large even when controlling for poverty levels and racial composition of the population (Buonanno 2003). In the US context, several papers �nd evidence linking the large increase in wage inequality to criminal behavior (Chiu and Madden 1998; Kelly 2000). One cross-country study found that higher inequality results in higher homicide rates (Fajnzylber, Lederman and Loayza 1998). Labor market conditions (formal job creation and destruction). Most studies linking labor markets to crime have used unemployment rates as a proxy for labor market conditions. However, this association is ambiguous. Although a large number of papers have found posi- tive association between crime rates and unemployment rates (Fleisher 1966; Ehrlich 1973; Grogger, Freeman 1994; Imrohoroglu et al. 2000), this association is at best weak. The weak relationship between unemployment and crime appears to be associated with two factors. First, as shown by Imrohoroglu, Merlo and Rupert (2001) most criminals are in fact employed CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY and only a smaller fraction is actually unemployed.2 Second, unemployment rates for youth and not total unemployment rates are the ones that appear to matter the most. Grogger (1998) explores the relationship between property crimes and wages. His results suggest that declining wages partially explain rising youth crime during the 1970s and 1980s in the US. In order to address these speci�c concerns, our estimates control for formal job creation and destruction, both for the total population, as well as for younger males. Contextual factors Urbanization and population density. A recent World Bank study found that a city’s growth rate appears to have a stronger relationship with homicide rates than the city’s size or den- sity. The annual rate of city population growth correlates positively with the homicide rate (r=0.27) and is statistically signi�cant (p<0.06) (WB 2010). Fast growing urban centers with weak labor markets are predisposing factors that breed urban criminality. With respect to urban density, there is no consensus in the literature on the direction of the e ect. Some studies found positive correlations (Schuessler 1962; Galle, Gove and McPherson 1972), while others found the opposite (Kvalseth (1977) and still others found non-signi�cant rela- tionships between the two variables (Freedman 1975). These studies indicate nonlinearities of the e ects of urbanization on crime. It is probable that if urbanization levels are low, an increase in the urban population share can decrease crime, as more collective policing can take place. But as urbanization levels increase, additional population pressures can generate congested living conditions that are likely environments for crime. Micro factors Demographic factors such as age and gender appear to be strongly related to criminal activ- ity. Because young males are more prone to violence, changes in the cohort size of young males will likely have an e ect on crime rates (Wilson and Herrnstein 1985; Freeman 1991, 1996; Grogger 1991, 1995, 1998). In addition, larger youth cohorts may also have a detrimen- tal e ect on legal labor market options of other youth; therefore, demography plays a key role in explaining crime. Education (high school dropout rates). A number of papers study the relationship between crime and education, stressing that education increases the returns to legal activities, rais- ing the opportunity cost of illegal behavior (Buonanno 2003). If the returns to education and other human capital are low, youth will be more inclined to drop out of school. Dropouts have 2 Imrohoroglu, Merlo and Rupert (2001) show that for the US about 79 percent of criminals are employed while only 21 percent are unemployed. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME the free time to engage in theft, the sale of illegal drugs, and other criminal activities (Becker 2007). Education may reduce the incentive to engage in criminal activities through several channels (Buonanno and Leonida 2005). First, schooling signi�cantly reduces the probability of incar- ceration (Lochner and Moretti 2003); higher levels of education are associated with higher returns in the labor market, which in turn increases the opportunity cost of engaging in il- legal activities. Second, education changes personal preferences a ecting the decision to engage in criminal behavior though a civilization e ect (Fajnzylber et al., 2002). Third, school enrolment alone, independent of the level of educational attainment, reduces the time avail- able for participating in criminal activity (Witte and Tauchen 1994). FACTORS THAT CORRELATE WITH NATIONAL TRENDS In order to understand how these factors correlate with national homicide rates, we base our analysis on �xed e ects regressions using data for all municipalities in Brazil, as well as for all �ve regions for 1998–2008. In addition to the regression analysis, we use Oaxaca-Blinder to decompose the contribution of endowments and coe cients to the crime reduction ob- served between 2003 and 2008. The analysis is based on an original database that contains information for the six risk factors and homicide rates for the 5,564 Brazilian municipalities for the 1998–2008 period. Estimates presented here are those reported in Calderón and Chioda, 2011. Homicides in Brazil decreased from 28.16 per 100,000 inhabitants in 2003 to 26.21 per 100,000 inhabit- ants in 2008, a reduction of 6.922% in the national homicide rate. Table 2: Elasticities of macro, contextual and micro factors and the homicide rate in Brazil 1998-2008 Correlation Changes in Factors 2003 v. 2008 Macro Factors 1% increase in the GDP +0.08 % The GDP increased 59% 1 percentage point increase in the value of the Gini Coefficient +1.01% The Gini coefficient declined in 3.13 percentage points 1% increase in the number of formal jobs created for males -0.049% Jobs created increased 65% 15-29 1% increase in the number of formal jobs destroyed for males +0.063% Jobs destroyed increased 67% 15-29 Context Factors 1 percentage point increase the Urban Population Share +10.52 % The Urban population share increased 2.88 percentage points CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY Correlation Changes in Factors 2003 v. 2008 Micro Factors 1% increase in the number of males 15-19 +0.40% The number of males 15-19 declined by 8.6% 1% increase in the number of males 20-29 +0.25% The number of males 20-29 increased by 13.07% 1 percentage point increase in the public school dropout rate +0.34% The public school high school dropout rate declined 2.69 percentage points Observations 19,121 R-squared 0.714 Note: All coefficients are significant at the 1 percent level. Source: Calderón and Chioda, 2011.3 From the analysis of the results presented in Table 2, it emerges that the GDP, the Gini coef- �cient, destruction of formal jobs for males aged 15-29, urbanization, the number of males 15-29 and 20-29 and the school dropout rate display robust positive correlations to the ho- micide rate. Formal job creation appears to be negatively correlated with homicide rates. All results have signi�cant coe cients, and tell us about the direction of the correlation of these factors and the homicide rate. We observe a positive correlation between the GDP and the homicide rate. Ehrlich (1973) identi�es GDP per capita and the growth rate of the GDP as proxies for the general level of prosperity in the provinces, as well as indicators of ille- gal income opportunities. While additional resources in an economy appear to suggest that there is more prosperity, if the gains of economic growth are not evenly distributed in the economy then people may take recourse to crime. The positive correlation we observe is a combination of opportunity v. additional resources, where opportunity appears to dominate. It is important therefore not only to study the correlation between the GDP and the homicide rates, but also to understand the correlation between inequality and crime. We �nd that the correlation of the GDP and the homicide rate is 0.08. In addition, our results suggest that the relationship between GDP and crime may be non-linear. The empirical results suggest that crime is positively correlated to GDP growth for low and medium levels of income. However, crime is negatively correlated to GDP for high levels of income, even after controlling for levels of inequality. Graph 6 below shows the relationship between log homicide rates and the log of the GDP. The non linearity of the relationship is not picked up in a simple regression that does not include polynomials. The graph shows that for lower income levels the relation- ship is positive, and it only becomes negative after a given threshold. 3 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Source: DATASUS, IBGE and RAIS, CAGED from 1997 to 2008 for all 5,564 municipalities. Information on school dropout rates is not available for 2006, and thus this year is dropped from the saturated model regressions. Year, Regional and State �xed e ects are included in the regressions, as well as the log of the population at the municipality level. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Graph 7. Scatter plot of the Homicide rate and the GDP in log scale 2 Weighted Log Homicide Rate per 100000 0 -2 -4 -6 -8 Log GDP Source: Calderón and Chioda, 2011. The observed relationship between inequality and crime also suggests an opportunity story: the more inequality, the more it would be pro�table for certain segments of the population to engage in criminal activity. The semi-elasticity of crime with respect to inequality measured by the Gini coe cient is (1.01) for Brazil between 1998 and 2008. This result implies that in areas where inequality is high, crime is likely to be high. Inequality coupled with restricted options, in particular for groups at risk, likely translates into more criminal behavior. We study the e ects of labor markets and crime using formal job creation and destruction for male youth. We �nd that formal job creation appears to be negatively correlated with the homicide rate (-0.049), whereas formal job destruction appears to be positively correlated with the homicide rate (+0.063). Weak labor markets appear to be a risk factor associated with crime and the more opportunities available to males 15-29 the more likely we will ob- serve lower levels of criminal behavior. Urbanization is the key context factor we study. We �nd that urbanization is positively cor- related with the homicide rate (10.52). Rapid urbanization processes are likely to involve poor migrant populations who �nd it hard to assimilate to their new urban environments and have di culty �nding jobs and licit opportunities at their destinations. However, we observe large non linearities in the relationship between crime and urbanization. For lower levels of urban- ization, an increase in the urban population share can reduce crime as the larger population can better police criminal activity. Yet as urbanization grows, congestion e ects take hold, increasing crime. The youth cohort of males 15-19 and 20-29 also appears to be positively correlated with crime. The larger the cohort, the lower the economic opportunities will be for each individual member, making it more attractive for them to engage in criminal activities as an outside CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY option. A series of preventative measures can be taken by policy makers to increase indi- vidual chances of labor market success. Stronger labor market opportunities for youth as well as policies that keep youth in school longer may prove successful in deterring criminal behavior. Our estimates suggest that school dropout rates are positively associated with the homicide rate, and thus e orts to keep kids and specially males 15 and above in schools should be undertaken to prevent crime. Dropout rates as well as the size of the cohorts of young males suggest that there is a clear positive correlation between these variables and violent crimes. Policies aimed at keeping youth at risk in schools (such as conditional cash transfer programs) can thus be used as a tool for crime prevention. We concentrate in this chapter on the link between high school dropout rates at the munici- pal level and the homicide rate. We believe that the correlation of these factors sheds light onto an important risk factor, namely unskilled male youth. The large correlations between homicide rates and the cohorts of school age males, as well as the high school dropout rate, suggests that policies targeting schooling and labor market opportunities for youth are likely to have large impacts on observed homicide rates. In chapter 4, this report further evaluates the e ects of two educational policies on crime. The importance of these factors serves as key motivation of the potential role of these types of policy interventions for youth. We estimate that incidence of violent crime appears to have some degree of persistence (CC, 2011). The results from the estimation suggest a sizeable inertia in homicide rates of about 30% in the �rst period that does not appear to vanish after six periods. Strong inertia of crime rates suggests that e orts to reduce crime are likely to be only moderately successful in the short run. Yet understanding the dynamics of crime is important if we are to learn more about risk and preventative measures that can be used in tackling current and future crime levels. Oaxaca Decomposition4 While the elasticities tell us how “sensitive� crime is with respect to a given factor in the 1998-2008 period, the Oaxaca Blinder decomposition (OB) provides us with descriptive sta- tistics to understand how changes in these factors explain the observed trends in homi- cides. The decomposition allows separating the observed change in homicide rates between 2003 and 2008 into changes in coe cients and endowments. Changes in coe cients refer to changes in correlations, or the marginal e ects of factors; while changes in endowments refer to changes in the mean values of these factors when comparing two separate periods. The procedure allows estimating how changes in the means of these factors as well as 4 The results from the Oaxaca Blinder decomposition presented here are not to be read as a causal interpreta- tion of the results, the factors we are controlling for may be endogenous. If in fact these factors are endog- enous we cannot establish if homicides are caused by changes in these factors, or if changes in these factors are produced by changes in the homicide rate. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME changes in the correlation coe cients, impact the total observed change in homicide rates between two periods, in our case 2003 and 2008. Graph 8: Decomposition of Changes Homicide rates for Brazil and its regions (with Changes Shown as % of Overall Increase) 300% 250% 200% 150% 100% Coefficients 50% Endowments 0% Brazil Southeast Northeast North South Central West -50% -100% -150% -200% Source: Calderón and Chioda, 2011. Graph 7 shows the Oaxaca Blinder decomposition for Brazil and its regions between 2003 and 2008. Changes in coe cients and changes in endowments add up to 100% of the ob- served change. For Brazil as a whole, 85% of the total change in homicides observed be- tween 2003 and 2008 is attributable to changes in endowments of the macro, contextual and micro factors, while 15% is attributable to changes in coe cients. The story is not homoge- neous across regions. For the Southeast and Northeast we �nd that changes in coe cients are relatively more important, while for the North, South and Central West regions, changes in endowments appear to be more relevant in explaining the change in homicide rates. In Graph 7, the sum of positive bars less negative bars represents the total change in homicide rates (100% of the change). The darker gray bars embody the fraction of the total change that is explained by changes in the endowments (e.g. changes in mean values of the six socio- economic factors), while the lighter gray bars denote the portion of the change that results from changes in the returns to the corresponding characteristic between 2003 and 2008. CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY Graph 9: Decomposition of Changes of Endowments in Factors (with Changes Shown as % of Overall change in endowments) 70% 60% 50% 40% 30% 20% 10% 0% -10% -20% Males 15-19 Males 20-29 Urban Population Share DP Dropout rate Gini Coefficient Males 15-29 Formal Jobs Added Males 15-29 Formal Jobs Destroyed Source: Calderón and Chioda, 2011. Changes in endowments refer to changes in mean values of these factors. Graph 8 shows that 58% of the total change in endowments is due to changes in the male cohorts of 15-19 year olds, while changes in the Gini coe cient account for 25% of the change in endowments and the high school dropout rate accounts for 11% of the total change in endowments; and the rest of the factors account for less than 10 percent of the change in endowments. The sum of positive bars and negative bars equals the total change in endowments (100%). These results imply that changes in the mean values of these three factors explain a large portion of the change in homicide rates that took place between 2003 and 2008. As we mentioned before changes in endowments account for 85% of the observed change in homicides at the national level, this implies that changes in the mean values of these three factors account for more than 80% in the total change in homicide rates observed between 2003 and 2008. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Graph 10: Decomposition of Changes of Characteristics in Factors (with Changes Shown as % of Overall change in characteristics) 250% 200% 150% 100% 50% 0% -50% -100% -150% Males 15-19 Males 20-29 Urban Population Share GDP Dropout rate Gini Coefficient Males 15-29 Formal Jobs Added Males 15-29 Formal Jobs Destroyed Source: Calderón and Chioda, 2011. Changes in coe cients refer to changes in the marginal e ects of factors or their relative importance. The sum of positive and negative bars adds up to the total change in coe cients (100%). Graph 9 shows that 213% of the total change in coe cients is due to changes in the male cohorts of 20-29 year olds, changes in the GDP account for 14% of the change in coe - cients. The rest of the factors account for less than 10 percent in the change in endowments. The main results are: š changes in the male cohorts of 15-19 year olds is the single most important predic- tive factor for the decline in homicide rates. It consistently accounts for large shares of the change in every homicide rate in Brazil between 2003 and 2008 (26.28% overall change in endowments). š changes the distribution of income (Gini coe cient) account for a second largest por- tion (25.72% of overall change in endowments). š changes in high school dropout rates appear as a third relevant factor in the reduc- tion of homicides at the national level (15% of the overall change). š Increases in urbanization, while predicting a reduction in homicides between 2003 and 2008, have a more modest e ect (1.23% of the overall change). š Factors such as the GDP growth, the increase in the male cohorts of 20-29 year olds, and the destruction of formal jobs for males appear, on the contrary, to have had a positive e ect on the homicide rate in this same time frame. Yet the overall e ect is a reduction in the homicide rate at the national level. CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY FACTORS THAT CORRELATE WITH CHANGES ACROSS AND WITHIN REGIONS Changes across and within regions As chapter 1 showed, the gradual decline in homicide rates at the national level masks the dramatic shifts that have been taking place across regions, states, and municipalities. We begin by presenting the panel regression estimates of the elasticities and semi-elastici- ties of crime with respect to the macro, context and micro factors at the regional level. As we show in chapter 1, the homicide rate only e ectively declined in the Southeast region after 2003, driving the national average down. While in the South, North and West Central regions the homicide rate remained largely unchanged, the homicide rate substantially increased in the Northeast. Table 3: Correlation Coefficients and changes in risk factors and homicide rate across regions (1998-2008) Central West Significance Significance Significance Significance Significance Significance Southeast Northeast region region region region region South Brazil North Macro Factors Log GDP 0.08 *** 0.08 *** 0.19 *** 0.24 *** 0.03 -0.01 Gini Coefficient 1.01 *** 2.63 0.87 3.69 0.39 -0.16 Log formal jobs created -0.05 *** -0.10 *** -0.02 0.02 -0.13 *** -0.06 for males 15-29 Log formal jobs destroyed for males 0.06 *** 0.02 0.07 *** 0.15 *** 0.04 0.05 15-29 Context Factors Urban Population Share 10.52 *** 22.75 1.18 -18.20 ** 23.33 ** 6.86 *** Micro Factors Log males 15-19 0.40 *** 0.49 *** -0.44 *** -0.70 ** 0.96 *** 0.31 ** Log males 20-29 0.25 *** 0.42 *** 0.39 *** 1.09 *** -0.30 ** 0.10 Public school dropout 0.34 *** 0.16 0.37 *** 0.39 0.10 0.13 rate Observations 19,121 6,802 5,318 874 4,304 1,823 Adjusted R-squared Source: Calderón and Chioda, 2011.5 5 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Source: DATASUS, IBGE and RAIS, CAGED from 1997 to 2008 for all 5,564 municipalities. Information on school dropout rates is not available for 2006, and thus this year is dropped from the saturated model regressions. Year and State �xed e ects are included in the regressions. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Panel estimates results In the Southeast the macro, contextual and micro factors appear to explain about 78% of the variation in the homicide rate, while in the other regions the proportion is much lower. In the Northeast in particular, we observe a statistically signi�cant increase in homicide rates after 2003, and the proportion of explained variation in homicides is 66%. Di erences in the pro- portion of the variation explained across regions suggest that factors not explicitly controlled for in the regressions are at least partially responsible for the di erences in homicide rate trends. We also observe di erences in the elasticities or correlations of these factors with respect to the homicide rate across regions. The elasticity of Homicide rate with respect to GDP ap- pears fairly large and positive i.e. an increase in the GDP can increase opportunity for crime, but at the same time it may imply less necessity for crime; while the two channels exist, the opportunity channel appears to dominate the necessity channel as we �nd a positive cor- relation between GDP and the homicide rate at the regional level. This correlation appears to be substantially larger in the Northeast (0.19) and the North (0.24) than in the rest of the country (0.08). The income Gini correlation coe cient appears to be particularly large for the North (3.69) and Southeast (2.63) regions, with correlations three and two times as large as those ob- served for the national average (1.01), respectively; these results are suggestive of a po- tential channel for crime reduction linked to concentration of income. Formal jobs created appear to be negatively correlated with crime rates across the board, yet with a statistically signi�cant and large e ect in the Southeast and in the South. This suggests that job creation for youth in these regions appears to be a potential key factor. Urbanization is strongly positively correlated with crime in the Southeast (22.75) and South (23.33) regions, moderately positively correlated with crime in the Northeast (1.8) and Central West regions (6.86), and negatively correlated with crime in the North region (-18.20). These results make sense as we observe larger shares of urban populations in the Southeast and the South, and an urbanization process taking place in the Northeast and Central West re- gions, while in the North the urban population share is relatively small. For initial stages of urbanization, the rising concentration of individuals in towns reduces crime as people are less exposed to individual risks. If urbanization increases rapidly, however, there is a conges- tion e ect that increases the likelihood of crime. These results are consistent with the levels of urbanization in the country and we �nd larger correlations in regions with a larger share of the population living in metropolitan areas. CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY The e ects of the changes in the cohorts of younger males (15-19) are also not uniform across regions. In the Southeast (0.49), these cohorts appear to have a larger positive cor- relation with the homicide rate than in the rest of the country, while we observe a negative correlation of the size of these cohorts in the Northeast (-0.44) and North (-0.7) regions. The story is di erent for the size of the cohorts of males 20-29. In the Southeast (0.42), North (1.09), and Northeast regions (0.39), these cohorts appear to have a large and positive cor- relation with the homicide rate, while they have a small and negative correlation with the homicide rates in the South, and no correlation in the Central-West region. Public school dropout rates appear to be positively correlated with the homicide rate both at the national and regional levels. The correlation, however, appears to be particularly large for the North (0.39) and Northeast (0.37) regions. The panel data analysis allows us to estimate elasticities and semi-elasticities by exploiting the within and between variation in crime rates at the municipality level. This type of analysis provides cleaner estimates of the elasticity or semi-elasticity of these macro, contextual and micro factors with respect to homicide rates. These results show clear di erences in the cor- relations of these factors to the homicide rate across regions. Crime in the North, South and Center-West regions increased but relatively less than in the Northeast. Table 4 shows changes in the macro, contextual and micro factors between 2003 and 2008. We observe that while GDP substantially increased in all regions, the Gini coe - cient declined, but to a lesser extent in the Northeast and Central-West regions, and net job creation for males 15-29 was positive in the Southeast and in the Northeast. Urbanization in percentage points increased in the Northeast, South and Central West regions. The cohorts of males 15-19 declined in all regions but more in the Southeast than in the other regions, and the cohorts of males 20-29 declined in this region while it substantially increased in all other regions. The high school dropout rates declined in all regions except the Northeast, where it increased by about one percentage point. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Table 4: Changes in macro, context and micro factors by region 2003 v. 2008 Homicide Rate Homicide rate per 100,000 2003 2008 Change Southeast 35.44 21.59 -39.08% Northeast 17.64 28.53 61.73% North 22.51 32.98 46.52% South 19.31 24.19 25.26% Central-West 29.47 30.77 4.43% Macro Factors GDP Gini Coefficient Jobs creaed males 15-29 Jobs destroyed males 15-29 2003 2008 Change 2003 2008 Change 2003 2008 Change 2003 2008 Change (Percentage points) Southeast 400,000,000 66,500,000 66% 0.55 0.51 -3.87 59,618 112,155 88% 50,618 91,864 81% Northeast 2,728,024 5,052,679 85% 0.58 0.55 -2.91 6,941 12,511 80% 5,956 10,111 70% North 3,862,960 7,149,995 85% 0.53 0.50 -3.02 7,231 13,980 93% 6,024 12,228 103% South 3,968,944 7,007,792 77% 0.53 0.49 -3.96 10,179 17,480 72% 8,952 15,372 72% Central-West 13,200,000 25,600,000 94% 0.56 0.54 -1.52 18,640 29,962 61% 16,273 25,606 57% Context Factors Urbanization 2003 2008 Change (Percentage points) Southeast 28.12% 27.99% -0.13% Northeast 30.55% 31.49% 0.93% North 38.60% 38.37% -0.23% South 11.49% 11.65% 0.15% Central-West 31.75% 32.17% 0.42% Micro Factors Males cohorts 15-29 Males cohorts 20-29 High School Dropout Rate 2003 2008 Change 2003 2008 Change 2003 2008 Change (Percentage points) 102,322 86,123 -15.83% 198,523 194,590 -1.98% 12.2% 9.2% -3.0% 22,945 22,003 -4.11% 37,993 47,539 25.13% 18.2% 19.0% 0.8% 22,866 21,445 -6.21% 39,622 43,706 10.30% 19.6% 18.4% -1.2% 13,652 12,833 -6.00% 25,041 27,666 10.48% 13.7% 11.3% -2.4% 31,950 31,779 -0.53% 60,368 67,652 12.07% 18.8% 15.1% -3.7% Source: Calderón and Chioda, 2011. Similarly to our national level analyses, we also report the Oaxaca Blinder decomposition for two of the regions, namely the Southeast and the Northeast. 6 Oaxaca Blinder decomposition for the Southeast and Northeast regions The Oaxaca-Blinder decomposition analysis complements the results from the panel regres- sions at the regional level by making it possible to determine which factors contributed to the reduction in the homicide rate in the Southeast and to the increase in the homicide rates in the Northeast in particular. The changes in endowments are the changes in the mean values of these factors, and the changes in coe cients refer to changes in correlations. Changes in coe cients are particu- 6 We only report the results for these two regions because they have divergent trends in the time frame of interest, but also because the availability of data for these regions allows performing the decomposition with no additional assumptions. CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY larly large for the Southeast and explain about 90 percent of the total change in homicides between 2003 and 2008 for this region. However, the Southeast follows the national trends in the importance of factors, i.e., the reduction in the younger male cohort of 15- to 19-year-olds and the Gini coe cients appear as the most prominent factors. In the Northeast, changes in endowments only represent -154 percent of the overall change; the change in coe cients appears to explain 254 percent of the observed change in this region. Here the changes in correlations appear to more than double the contribution of changes in endowments; these changes in correlations mostly explain the observed change between 2003 and 2008 in the homicide rate in this region. Similar to the Northeast, in the Southeast the bulk of the di erence is explained by changes in the marginal e ects, rather than by changes in means. However, for the Southeast most of the observed di erences in homicide rates between 2003 and 2008 are due to changes in coe cients (marginal e ects) of the youth cohort of males 15 to 19 years of age, changes in the concentration of income, and the dropout rate. The endowments and coe cients associ- ated with the decomposition are large for the cohorts of males 15-29 in Brazil as well as in the Southeast and the Northeast regions. We decide to analyze graphically the contribution of the other factors to the changes in endowments and coe cients. Graph 10 and 11 show the contribution of the GDP, the Gini coe cient, job creation and de- struction, urbanization and high school dropout rates to total endowments and coe cients for Brazil, the Southeast region, and the Northeast region. As illustrated in these graphs, the bars depict the total size of the e ect of endowments and coe cients, respectively. Graph 11. Contributions of endowments for Brazil, the Southeast region, and the Northeast region 2008 v.2003 1.0 0.8 0.6 Jobs detroyed for males 15-99 0.4 Jobs Added for males 15-99 Gini coefficient 0.2 High School Dropout rate 0.0 Log GDP -0.2 Urban Population Share -0.4 -0.6 Source: Calderón and Chioda, 2011. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Graph 12: Contribution of coefficients for Brazil, the Southeast region, and the Northeast region 2008 v.2003 5.0 0.0 Jobs detroyed for males 15-99 Jobs Added for males 15-99 -5.0 Gini coefficient High School Dropout rate -10.0 Log GDP Urban Population Share -15.0 -20.0 Source: Calderón and Chioda, 2011. For the Northeast, the prominent factors are the changes in marginal e ects associated with the Gini coe cient, the dropout rate and urbanization. First, changes in marginal values of urbanization in this region appear to have a more negative e ect than in the Southeast and Brazil as a whole in this period. Second, the larger marginal e ect of dropout rates in this region suggests a higher propensity toward violent crime among dropouts in this region in particular. Policies targeted to keep youth at risk in school should be implemented as crime prevention measures, in particular in the Northeast region, where the change in propensity to crime associated with this factor is large. Third, the larger marginal e ect of the Gini coef- �cient also suggests that policies which reduce the concentration of income in the hands of a few will be a particularly e ective preventive factor against violent crime. The main results are: š Although at the national level the proportion of the variation in homicide rates ex- plained by the macro, context and micro factors described in this chapter is about 71 percent, it varies substantially at the regional level, explaining about 78 percent of the variation in the homicide rate in the Southeast and about 66 percent in the Northeast. š The correlation of the Gini coe cient and the homicide rate is 2 and 3 times larger in the Southeast and North, respectively. š The correlation between urban population share and the homicide rate is positive for all regions except the North, which is the region with the lowest percent of urban population. The correlations are particularly large for the Southeast and the North and substantially smaller for the Northeast and the Center-West regions. š The cohorts of younger males (ages 15 to 19) appear to be negatively correlated with crime in the Northeast and positively correlated with crime in all other regions. The cohort of males ages 20 to 29 is positively correlated with the homicide rates for all regions. CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY š The dropout rate appears to be positively correlated with the homicide rate, but it is only statistically signi�cant for the Northeast, which has a correlation three times larger than that observed at the national level. š The Oaxaca decomposition of the Southeast and Northeast gives us a clearer idea of what changes took place between 2003 and 2008 in these regions. We �nd that for both regions, changes in coe cients appear to matter most, in particular in the Northeast. However, we �nd some striking di erences on how changes in these mar- ginal e ects a ected the di erences in outcomes. š Over time, we observe a larger marginal e ect of high school dropout rates, the concentration of income, and urbanization rates in the Northeast, with coe cients more than twice as large as those observed at the national level. š In the Southeast, most of the observed di erences in homicide rates between 2003 and 2008 are due to changes in coe cients (marginal e ects) of the youth cohort of males 15 to 19 years of age, changes in the concentration of income, and the high school dropout rate. When compared to the Northeast, we observe a smaller change in these coe cients, implying a larger marginal change in the Northeast relative to the Southeast or the national average. Although the changes in these factors suggest that di erences in endowments and coef- �cients account largely for the di erent regional outcomes, a large fraction of the variation is not explained by these factors, and could potentially be explained by variables that are in principle controllable by policies. We present some suggestive evidence of these controllable factors in the following section. ANALYSIS OF THE “RESIDUALS�—A POTENTIAL ROLE FOR POLICY? The factors considered in the previous section deliberately did not include proxies for tar- geted government actions/policies aimed at reducing crime incidence. The observed trends in crime are thus decomposed in two parts: one that is explained by the macro, context and micro factors, and the other that is explained by all the factors we did not include in the regression, including policy actions. In light of this mechanical decomposition, the residual component can be interpreted as a noisy measure of the role of policies. It is important to stress that the residual component can at best only be correlated with policy action, because of our inability to control exhaustively for all those factors other than policy that could influ- ence crime trends, for instance migration and displacements, endogeneity, etc. Therefore, the evidence presented in this section is at best suggestive of the role of poli- cy, and so particular caution and thoughtfulness are needed in interpreting these results. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME š Larger positive overall residuals can be an indication of events or policies that posi- tively a ect the crime rates, while lower or negative overall residuals could suggest that a number of unobserved factors not controlled for explicitly in the regression are helping to explain a reduction in violent crime. š Our estimates �nd large heterogeneity in all of these residuals across and within regions and states. š The frequency-weighted means of these residuals suggest that we will observe more positive residuals, in particular in the North and Northeast, and more negative residu- als in some municipalities in the Southeast. The results are not conclusive, because we �nd municipalities from all regions in the top 100 positive residuals, and munici- palities from all regions in the top 100 negative residuals. š We observe that larger positive residuals in municipalities in the Southeast corre- spond to areas that are not part of the metropolitan areas in the state. Lower (nega- tive) residuals in municipalities in the Southeast correspond to municipalities in metropolitan areas, in particular those of Belo Horizonte and São Paulo. Table 5: Frequency-Weighted Largest Positive 100 Overall Residuals (after fixed effects estimation, 1997–2008) Region State Freq. Percent North PARA 22 22% South PARANA 13 13% Southeast MINAS GERAIS 11 11% Southeast SAO PAULO 10 10% Northeast BAHIA 9 9% Northeast CEARA 6 6% South RIO GRANDE DO SUL 6 6% Southeast ESPIRITO SANTO 5 5% Center-West MATO GROSSO 3 3% Northeast PARAIBA 3 3% Northeast PIAUI 3 3% Northeast RIO GRANDE DO NORTE 3 3% Center-West GOIAS 2 2% Northeast ALAGOAS 1 1% Northeast MARANHAO 1 1% Northeast SERGIPE 1 1% Southeast RIO DE JANEIRO 1 1% Total 100 100% Source: Authors’ calculations based on DATASUS, IBGE and RAIS, CAGED from 1997 to 2008 for all 5,564 municipalities. Information on school dropout rates is not available for 2006, and thus this year is dropped from the saturated model regressions. CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY Table 6: Frequency-Weighted Largest Negative 100 Overall Residuals (after fixed effects estimation, 1998–2008) Region State Freq. Percent Southeast MINAS GERAIS 25 25% North PARA 17 17% Southeast SAO PAULO 13 13% Northeast BAHIA 12 12% Northeast GOIAS 11 11% Center-West MATO GROSSO 5 5% South PARANA 5 5% Northeast CEARA 3 3% Southeast ESPIRITO SANTO 3 3% Northeast MARANHAO 3 3% Northeast PARAIBA 1 1% South RIO GRANDE DO SUL 1 1% Northeast SERGIPE 1 1% Total 100 100% Source: Authors’ calculations based on DATASUS, IBGE and RAIS, CAGED from 1997 to 2008 for all 5,564 municipalities. Information on school dropout rates is not available for 2006, and thus this year is dropped from the saturated model regressions. While the evidence presented here is only suggestive, one can associate these positive re- siduals with unobserved factors that are increasing the homicide rate, while we can asso- ciate the negative residuals with unobserved factors that reduce the homicide rate. While table 5 and 6 present evidence on the 100 largest positive and largest 100 negative residuals respectively, 7 shows the ranking for the overall means, median, 25th percentile and the 75th percentile. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Table 7: Residuals ranking (negative to positive) Ranking State Region Residual Residual Median Residual Percentile 25 Residual Percentile 75 Amazonas North 1 18 16 6 Rio Grande do Norte Northeast 2 16 8 4 Piaui Northeast 3 14 13 3 Santa Catarina South 4 1 19 1 Bahia Northeast 5 9 3 16 Sergipe Northeast 6 13 15 12 Mato Grosso Central West 7 7 5 18 Sao Paulo Southeast 8 5 7 13 Ceara Northeast 9 15 2 17 Goias Central West 10 12 12 11 Rio Grande do Sul South 11 4 17 5 Espirato Santo Southeast 12 17 9 8 Para North 13 8 1 19 Alagoas Northeast 14 11 11 15 Paraiba Northeast 15 10 10 7 Parana South 16 3 6 9 Minas Gerais Southeast 17 2 14 10 Maranhao Northeast 18 6 4 14 Rio de Janeiro Southeast 19 19 18 2 Residual estimates were calculated after fixed effects estimations of the saturated model. Residuals were ranked using state frequencies. Source: Calderón and Chioda, 2011. Negative residuals imply that the observed homicide rates are below the predicted homicide rates, and positive residuals imply that predicted homicide rates are above the predicted ho- micide rate. We provide a description of the means following table 7. In the Southeast region, we observe that Rio de Janeiro and Minas Gerais have positions 19 and 17 respectively in this ranking. This implies these two states have the most negative, and the third most negative residuals in the sample (out of 19 states). In this ranking Sao Paulo is ranked as number 8, and Espirito Santo as number 11. In the Northeast region, Rio Grande do Norte ranks 2, Piaui 3, Bahia 5, Sergipe 6, Ceara 9, Alagoas 14, and Maranhao 18. In the North, Amazonas ranks 1, and Para ranks 13. In the South region, Santa Catarina ranks 4, Rio Grande do Sul ranks 11, and Parana ranks 16. In the Central West region, Matto Grosso ranks 7 and Goias ranks 10. We observe four states (Rio Grande do Norte, Piaui, Bahia, and Sergipe) in the Northeast with a more positive residual than the �rst state in the Southeast (Sao Paulo). Only one state in the Northeast has a negative residual (Paraiba), whereas we �nd two of the states in the South- east have negative residuals (Minas Gerais and Rio de Janeiro). These rankings show that there are a larger number of states in the Northeast that are part of the top �ve residuals by size (most positive), while there are two states in the southeast CHAPTER 2. REDUCING CRIME IN BRAZIL: THE FACTORS DRIVING THE CHANGE AND REGIONAL HETEROGENEITY in the top �ve for negative mean residuals. Policies captured by these residuals could po- tentially explained some of the variation in homicide rates in these regions, as we observe a marked increase in homicide rates in the Northeast and a sharp decline in the Southeast following 2003. CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME This chapter provides an overview of the citizen security policies implemented by some of the top-performing subnational governments in Brazil: the State Governments and capi- tals of São Paulo, Minas Gerais and Rio de Janeiro. As Chapter 2 suggests, a significant part of the decline in violent crime in these states is strongly correlated with factors that are not directly related to sub-national government policies, such as the reduction of the youth male cohort, or reductions in inequality. However, the forgoing analysis of “residuals� also suggests a potential role for the policies implemented by these states. This chapter reviews some of the evidence on the potential contribution of these policies to the decline in violent crime and the lessons that could be drawn from them. The nationwide drop in violent crime since 2003 has been largely driven by steep de- clines within Brazil’s three most populous states and their respective capital cities, all of them located in the modern, industrialized Southeastern region. Although Minas Gerais did not perform as well the other two, the state was included in the analysis because of the sharp decline in the capital city of Belo Horizonte and its metropolitan region (Graph 10). Graph 13. Homicide Rates for State and State Capital of São Paulo, Rio de Janeiro, and Minas Gerais, 1998-2008 80.00 70.00 60.00 Minas Gerais Homicide Rate per 100,000 50.00 Belo Horizonte Rio de Janeiro 40.00 Rio de Janeiro (city) São Paulo 30.00 São Paulo (city) 20.00 10.00 - 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 There are important similarities and di erences in the characteristics of the crime drop across the three cases: š In terms of size of the change, São Paulo shows the largest drop of the three. Its municipality went from being the second most violent of the 28 state capitals to the second least violent between 1999-2008 (76 percent decline from 61.1 to 14.8 homi- cides per 100,000). The State of São Paulo went from being the second most violent to the third least violent in the country, with its homicide rate falling 66 percent, from CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? 44.1 to 14.9 per 100,000 over that period. The state of Rio de Janeiro was next with a 60 percent drop between 2002-2008 (from 56.5 to 34 per 100,000) and the city of Rio de Janeiro with a 50 percent drop in the same period (from 62.4 to 30.9 per 100,000). While the decline in the state of Minas Gerais was small (5.3 percent between 2003- 2008), its capital of Belo Horizonte had a drop of 32.5 percent. š In terms of timing, São Paulo was a precursor to the change in the other states and the country overall. The crime drop began in 1999 in São Paulo, while in Rio de Ja- neiro only in 2003, and in Minas Gerais not until 2004/2005. š In terms of spatial coverage, all three cases show higher drops in their capital cities than in the rest of the state. In São Paulo the drop was more spread out across the state (a ecting 500 out of 645 municipalities) than in Minas Gerais and Rio de Janeiro where it was more focused in the capital. š In terms of the evolution of di erent types of violent crime other than homicide, São Paulo saw a decline among car robberies but an increase among thefts and bodily injuries. In Minas Gerais, the rate of violent crimes (including robbery, bodily injury, homicide, homicide attempt, rape, and rape attempt) decreased 45 percent in the state of Minas Gerais, and property crimes decreased by 45 percent between 2003-2009. In Rio de Janeiro, the crime drop has been largely restricted to homicides and, to a lesser extent, all intentional lethal crimes. Total property crime indicators have only been improving since 2008; before that, they actually worsened. Rates of theft and larceny have been on the rise state-wide since 2004, plateauing but not yet entering into decline even in the Capital City as late as 2010. Multiple, concurrent factors appear to have contributed to falling crime rates in all three states. In addition to the structural factors identi�ed in the previous chapter, the Brazil- ian literature and our own analysis suggest that speci�c policies have probably played an important role in the crime drop. The literature has identi�ed a wide range of potential sus- pects, from programs to control guns and alcohol consumption, to e orts to improve police e ciency, increased incarceration, increased spending on public security, increased social spending, etc. It is di cult to identify one single factor as the cause of the crime decline in any of the three states. This is consistent with the conclusions of studies that attempt to explain the crime shift in the USA, and particularly in New York City, as well as in Canada during the 1990s (see Blumstein and Wallman 2006; Zimring 2007). Instead, the explanation seems to lie in a number of factors, perhaps none of which would have been su cient to explain the drop in crime. These states and capitals have been at the forefront of innovative citizen security poli- cies in Brazil. These three states stand out not only because of their substantial crime drop but also because of the innovative policies that they have implemented to improve citizen security. A review of the evidence on the policies implemented in these states suggests the importance of four types of policy interventions: MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME š Strengthening the results orientation and managerial capacity of the police and pub- lic security system. š Reducing the presence of key environmental risk factors, particularly guns but also alcohol and unsafe urban spaces. š Targeting territories and populations at risk through multi-sectoral prevention strate- gies. š Integrated and multi-stakeholder subnational citizen security plans. The rest of the chapter is organized along these four types of interventions. The purpose is not to be comprehensive but to give a quick summary of the main evidence produced on the types of policies claimed by the literature to have made a di erence in reducing violent crime in these states and cities. It is important to keep in mind that the evidence basis on the e ectiveness of most of these policies is not strong enough to establish causal relationships; even when it is, the external validity or generalizability of any given policy is not established. At the same time, there is evidence for the e ectiveness of similar types of interventions in developed countries. These interventions correspond to the three most common and com- plementary approaches to deal with crime and violence: criminal justice, public health fo- cused on risks, and cross-sectoral approaches. REFORMING THE POLICE AND THE PUBLIC SECURITY SYSTEM Poor performance by the criminal justice system, and particularly by the police, has been identi�ed as one of the key systemic roadblocks to citizen security improvements in Bra- zil (Veloso and Guimarães Ferreira 2008). Some of the key factors that have contributed to this situation include: (a) a culture of reactive policing that is concerned with responding to each incident rather than identifying crime trends and preventing future incidents (World Bank 2006), which is partly related to (b) the lack of information and management systems to learn about crime patterns, and (c) the existence of two police forces (one responsible for ostensive patrolling—Military Police—and another one responsible for crime investigation –Civil Police) that lacked coordination, information sharing, and trust. This section highlights three innovations promoted by São Paulo and Minas Gerais to con- front these challenges. São Paulo led Brazil in adopting state-of-the-art, results-oriented public security management systems and providing policy makers with high-quality infor- mation for decision making. Minas Gerais followed São Paulo’s lead and established a con- solidated results-oriented management system for all law enforcement and public security agencies operating in the state. In Rio de Janeiro, progress in this area has been more recent but is moving in a similar direction, with the highlight of the reform being the introduction of Police Paci�cation Units (Unidades de Polícia Paci�cadora, UPP), presented in the third sec- tion of this chapter. CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? Results-Oriented Police Reforms in São Paulo. Toward the end of the 1990s, as a response to the dramatic crime increase in São Paulo, the State Government began a comprehensive set of reforms to introduce a results orientation in its police force. Although there is no ro- bust evaluation of these policies, strong anecdotal evidence has led many observers to credit them with having an important e ect on São Paulo’s drop in crime. These consisted of a broad set of changes, including the introduction of performance assessments; the territo- rial coordination of military and civil police forces; and community policing (Lima 2010). Key goals and initiatives for strengthening the results orientation of the police included: š Gathering accurate and timely empirical data and using it to plan and evaluate programs. An intergovernmental communications network was established to link the military and civilian police. The Crime Information System (Sistema de Informação Criminal, Infocrim) linked 93 police districts and 79 Military Police battalions and allowed po- lice to create strategies for 39 municipalities in the São Paulo Metropolitan Region (Região Metropolitana de São Paulo, RMSP). It began operating in 2000 along with Co- pom Online, an online database allowing for real-time monitoring of emergency calls. Fotocrim, which began in 2002, was a database with 207,903 photographs of people caught committing crimes and of prisoners (Waisel�sz and Athias 2005). š Capturing multiple-homicide perpetrators. In 2001, the government restructured its Homicide Department to give it a more strategic orientation. It created an Intelligence Analysis Service within the department and developed a Plan to Combat Homicide focused on “hardened killers,� or perpetrators of multiple homicides. As a result, the police were able to solve 65 percent of homicides (compared to an average of 7 per- cent for Brazil) and 93 percent of multiple-homicide cases in that year (Veloso and Ferreira 2008). š Engaging citizens in reporting crime. The “Disque Denúncia� Crime Hotline, rolled out in 2000, provided citizens with a free channel to anonymously report crimes or pro- vide tips to the police, and contributed to the police’s ability to prevent and resolve crimes. Created by the São Paulo Institute Against Violence (Instituto São Paulo Contra a Violência), Disque Denúncia received more than 1.6 million calls and 300,000 formal complaints between 2000 and 2004 in the São Paulo Metropolitan Region. It led to 13,000 successful police actions: 5,760 homicide cases solved; 82 percent of all mul- tiple homicides cases solved; 48 kidnapping victims freed; 9,420 arrests of criminals caught in the act; 1,965 arms and munitions seized; and 1,647 vehicles recovered (Kahn 2005; Disque Denúncia/ISPCV/SSP). Integrated Management of Public Safety (Integração e Gestão da Segurança Pública, IGESP) in Minas Gerais. The main objective of IGESP was to allow the free and immediate flow of information between the Military and Civilian Police forces and the coordination of integrated planning within the public safety community. The lack of coordination, information sharing, and trust between the Civilian and Military Police have posed problems in many Brazilian MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME states and in the country as a whole. IGESP introduced modern technologies for information monitoring and targeted policing, using a dynamic updating system and constant evaluation of strategies and actions by the police organizations. The program was implemented in Belo Horizonte in 2005 and subsequently expanded to 56 municipalities by 2008. IGESP became the backbone of public security policy in Minas Gerais for improved management and the monitoring and evaluation of public strategies to control and prevent crime and violence in the state (Cruz, 2005). Its adoption in the state was part of a more comprehensive and results-oriented public security strategy launched by the State Government in 2003 (see section below on integrated strategies). An impact evaluation of IGESP showed it had a causal e ect on crime reduction (Box 1.) IGESP consisted of �ve key reforms: š Data consolidation and management. IGESP consolidated all of the data and intelli- gence gathered by di erent police forces, including calls to emergency telephone lines and police vehicle dispatches, through an integrated information system. š Streamlined decision making. The program consolidated and streamlined decision making at the level of the Integrated Area of Public Security (�reas Integradas de Segurança Pública, AISP), through the territorial coordination of strategic actions in- volving the various law enforcement agencies. Under IGESP in Belo Horizonte, for example, Civil and Military police from certain AISPs began meeting weekly to de�ne strategic plans and objectives, which facilitated the integration of these two police forces. š Spatial analysis. The program adopted a georeferenced system to identify crime hotspots and target police responses. š Results-based management. The program established performance indicators and state-level targets for all law enforcement agencies, consolidating a results-oriented approach that improved police performance. š Improved police training. Finally, police training was intensi�ed and modi�ed to in- clude subjects such as human rights education, and more adequate incentives were introduced, including more generous pay raises and promotions. Box 1. IGESP Impact Assessment Provides Evidence of Strong Effect on Crime A recent evaluation estimated the program’s impact by applying a di erence-in-di er- ences strategy that exploited the staggered adoption of the program by municipalities and the evolution of the variables of interest before and after implementation. The study found a causal e ect of IGESP on crime. The most conservative estimates indicate a 24 percent reduction in property crimes and 13 percent in personal crimes. There is also CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? evidence that IGESP is associated with improved police response to crime, measured by the number of weapons apprehended and clearance rates. It found a negative but non-signi�cant e ect on homicides (Soares and Viveiros 2010). Rio de Janeiro. For most of the past three decades, several factors have prevented the suc- cessful implementation of a comprehensive statewide public security strategy along the lines of those pioneered by São Paulo and Minas Gerais. This dynamic has improved over the past several years, among other things by the increased coordination among city, state and Federal Government agencies with regard to public security policy in Rio. The police in Rio de Janeiro have created a system similar to the IGESP model, establishing the integration of police forces at the regional (Regiões Integradas de Segurança Pública, RISPs), area (�reas Integradas de Segurança Pública, AISPs), and circuit (Circunscrições Integradas de Segurança Pública, CISPs) levels and a performance system that rewards results. Targets are therefore broken down by city, neighborhood and police unit, and evaluated based on a set of indica- tors that includes number of car thefts, homicides, robbery and bodily injury. A new indicator, death by confrontation with the police (“autos de resistência�) is also being added to the sys- tem, a signi�cant progressive step, since in previous decades policemen had been rewarded for killing criminals. The results are tracked monthly, and if targets are met, police o cers can receive a bonus that can reach three times the salary of a soldier. When the results fall behind, an action plan is mandatory, but o cers are not punished (Alzir and Cesário de Sá 2011). This results-management approach is fairly recent in Rio (2009). There are no rigor- ous evaluations of its impact to date. The innovative managerial and information collection and sharing model adopted in the three states followed a trend inspired by the COMPSTAT model pioneered by the New York City Police Department (NYCPD) in the mid-1990s, which involved new “management for re- sults� techniques that changed the way performance information was used to improve police action.7 The system is also in use in other major cities in the US and other developed coun- tries including Canada, Australia and New Zealand that have witnessed signi�cant declines in crime rates. In a overview of the eight key police policies that could explain the decline in crime in NYC, which represent since mid 1990s the largest and longest sustained drop in 7 COMPSTAT, short for COMPuter STATistics or COMParative STATistics, is the name given to NYCPD’s account- ability process and has since been replicated in many other departments. COMSTAT is a management phi- losophy or organizational management tool for police departments. It is a multilayered dynamic approach to crime reduction, quality of life improvement, and personnel and resource management. It employs Geo- graphic Information Systems and was intended to map crime and identify problems. In weekly meetings, ranking NYPD executives meet with local precinct commanders from one of the eight patrol boroughs in New York City to discuss the problems. They devise strategies and tactics to solve problems, reduce crime, and ultimately improve quality of life in their assigned area. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME street crime ever experienced by a big city in the developed world, Zimring (2011) consid- ers management and data mapping COMPSTAT system as one of the major shifts that was probably successful, but acknowledges that speci�c e ects due to this policy can’t be proven. According to Kelling and Souza (2001), the impacts of this initiative are one of the most dif- �cult to statically measure because they are based on the “problem-solving� premise that police tactics need to be flexible to speci�c crime related problems as they are identi�ed via the Compstat system. Hence, both problems and tactics are constantly changing, which makes it di cult to establish a single measure of police activity that can be used as a proxy. Furthermore, in order to be e ective the system needs to be linked to the reforms, such as those followed in the three Brazilian states and described in the next sections (Bratton and Andrews 2010). REDUCING ENVIRONMENTAL RISKS: GUNS, ALCOHOL, AND UNSAFE URBAN SPACES The availability of guns and alcohol are two important risk factors associated with violence. Policies to curb their influence are credited in part for the decline in crime in São Paulo and Rio de Janeiro. The most compelling evidence stems from the e ect of gun control in São Paulo. Gun Control in São Paulo and Rio de Janeiro The risk of dying from wounds caused by guns is 2.6 times higher in Brazil than in any coun- try in the world (Dreyfus and Purcena 2009), and about 70 percent of homicides in Brazil are committed with �rearms (Ministry of Health 2008). The alarming growth in the percentage of deaths caused by �rearms led the Federal Government to enact gun control legislation in October 2003. These laws limited the imports of �rearms, made it illegal to own unregistered guns or to carry guns on the street, and increased the penalties for violation of gun control laws. Box 2. The Impact of Guns on Youth The externalities of armed violence mean that guns can have an impact on chil- dren’s welfare even if they never touch a weapon. A recent paper �nds a statistically signi�cant negative relationship between local rates of armed violence and school at- tendance as well as student performance on standardized mathematics exams in a sample of Rio de Janeiro public schools. The same study found a positive relationship CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? between local armed violence and grade repetition and dropout rates, with this e ect stronger among non-white students. The authors �nd evidence linking local violence to poor school attendance and performance, through direct channels such as teacher absenteeism and school closures, as well as through less direct channels, whereby high levels of local violence a ect household welfare even during periods of school vacation (Monteiro and Rocha 2010). Four of every ten apprehended arms in Brazil are in possession of youth. There is evidence to suggest that gun control policies implemented by the State and Mu- nicipality of São Paulo since 1997 have played an important role in the decline of homicides, and that these e ects were further ampli�ed by the e ective implementation at the state level of the 2003 federal Disarmament Law (Estatuto do Desarmamento). Between 1997 and 2002, the police increased the number of arms apprehended from 7,000 to 9,000 arms per trimester. This resulted from a 600 percent increase in police searches in the state between 1997 and 2006 (Kahn 2007). The Disarmament Law and the 2004 subsequent gun buyback program were evaluated by Souza et al. (2007). Both studies found positive results. São Paulo is considered to be an example at the national level because of its strict enforcement of the law (Souza et al. 2007). The state accounted for 25 percent of the nationwide decline in �re- arms deaths (Kahn 2010). Two recent studies have used more disaggregated data and stronger econometric tech- niques. One of them found that gun control policies in the state can explain 21 percent of the decline in homicides in the Municipality of São Paulo. It further calculated that one life was saved for every 18 weapons apprehended; a total of 13,000 lives were saved between 2001 and 2007. To achieve these results, the authors used the number of weapons apprehended as a proxy for the flow of guns, and the number of suicides per �rearm as a proxy for the stock of guns for 647 municipalities in the state between 2001 and 2007, in a �xed-e ects panel model analysis (Cerqueira and Mello 2010). The other study found smaller but similar results, speci�cally that a 60 percent reduction in the stock of weapons in the state between 2001 and 2007 led to a 9 to 12 percent decline in homicides over the same period (Hartung 2009). In Rio de Janeiro, the 2003–2004 Federal Gun Control Legislation and Small Arms Buy- back Campaign have been associated with drops in rates of armed violence statewide. A study commissioned by the United Nations Small Arms Survey and carried out by Brazilian researchers at ISER/Viva Rio in Rio de Janeiro found a 12 percent drop in �rearm-related deaths nationwide and an 11 percent drop in Rio de Janeiro attributable to the ban on carry- ing small arms and the accompanying 18-month-long handgun buyback campaign (Dreyfus et al. 2008). The timing of the sharp onset of the crime drop in Rio, beginning in 2003, coin- cides with the passing of nationwide gun control legislation that same year, accompanied by MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME aggressive e orts at the state and local levels by government and civil society in support of a successful small arms buyback campaign. The Brazilian evidence of the impact of disarmament and stricter gun control laws on crime reduction is widely accepted in the country. In other places where carrying arms is legally easier and culturally more accepted, such as the United States, gun control policies have been historically controversial, with influential authors and works providing evidence both in favor (Cook and Ludwig 2002) and against (Leott 2004) them. However, some of the most successful and internationally known interventions in the country, such as the pioneer “Cease�re Operation� in Boston, Massachusetts, which was replicated in other major cities like Chicago, had direct law-enforcement attack on illicit �rearms tra ckers supplying youth with guns as one of its key elements. A time series analyses of this program suggest a 63 percent decrease in the monthly number of youth homicides in Boston; 32 percent decrease in the monthly number of citywide reports of gunshots �red; and 25 percent decrease in the monthly number of citywide gun assault incidents (Braga et. al. 2000). A similar interven- tion that involved the con�scation of illegal arms in Kansas City also yielded a 49 percent decrease in �rearm-related crime. In Colombia, an evaluation of similar programs in Cali and Bogotá showed a decrease of 14 and 13 percent, respectively, in homicide rates on a week- end when the ban was enforced, compared to another one when it was not (Villaveces et al. 2000, in Cunningham et. all 2008). Regulating Alcohol Consumption through Bar Closures and Road Blocks The Dry Law (Lei Seca) was implemented in São Paulo Metropolitan Region between 2001 and 2004. The law consisted mainly of bar closures between 11 p.m. and 6 a.m. in order to limit the sale and consumption of alcohol, and was implemented by 16 di erent municipali- ties (Waisel�sz and Athias 2005). According to the di erence-in-di erence model used by Biderman et al. (2010), the 10 percent of the drop in homicides in the São Paulo Metropolitan Region could be attributed to the law, and similar results were found for other crimes such as battery. Using a causal model and intervention analysis, Kahn (2005) found a decrease in homicides in 6 of the 16 cities that had adopted the Dry Law by 2004. However, he points out that homicide also rose in some places that had adopted the law, and concludes that, al- though the Dry Law helped deepen the decline in homicide in the places where it declined, it cannot be considered su cient as a causal factor. The municipality of Diadema became the best-known example of all the cities that implemented the Dry Law in São Paulo [see Box 6]. In recent years Rio de Janeiro has also adopted a Dry Law, which is enforced with zero toler- ance and relative transparency at police road blocks, or “blitzes.� Any alcohol detected by the on-the-spot breathalyzer test administered to motorists suspected of being inebriated is suf- �cient to result in an immediate loss of driving privileges. And unlike in the past, when Rio’s CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? police o cers might be expected to negotiate informal solutions with motorists stopped for moving violations, no such discussion is permitted at Dry Law roadblocks, which are some- times attended by journalists from Rio’s local television and print media outlets. Improving the Safety of Urban Spaces During the great crime drop in São Paulo, parts of the city experienced an important pro- cess of slum upgrading and urban renewal (Cities Alliance 2009, Peres et al 2010). A similar process happened in Belo Horizonte and is currently taking place in Rio de Janeiro. In recent years, the municipal government of São Paulo has devoted signi�cant resources to improv- ing the urban environment and reducing urban poverty.8 These major urban improvements include the regularization and upgrading of slums under the Bairro Legal program, as well as the resettlement of households living in hazardous areas or the rehabilitation of the city center, which was one of the city’s crime hotspots (Martin and Jenkins 2008). An important strand of crime theory posits that there is a direct association between physical disorder in a neighborhood and crime. The so-called ‘broken windows’ theory ar- gues that disorderly and petty criminal behavior, such as the presence of abandoned and derelict buildings, very noisy, and poorly lit, may lead to more criminal activities (Kelling and Cole 1998). The implementation of responses to this theory in certain places, such as NYC, where it was originally ‘tested’, coincided with declines in rates of di erent crimes, inspir- ing a series of empirical studies supporting this type of policy.9 Thus, it is not surprising that some scholars have identi�ed urban renewal processes as one of the factors that have contributed to reduce the risk of victimization in São Paulo and Belo Horizonte (Peres et. al 2010, Peixoto 2010). While we were not able to test this hypothesis, we were able to estimate whether there is a relationship between neighborhood disorder and crime in São Paulo. Re- sults are summarized in Box 3 and in (Alda 2011). 8 Some of the most important programs include Programa Manaciais, Programa de Urbanização de Favelas, Programa de Regularização Fundiária em Favelas, Programa de Regularização e Urbanização de Loteamen- tos Irregulares, and the Programa de Mutirões. 9 Using misdemeanor arrests as a measure of the broken-windows approach, Corman and Mocan (2002) found that the model explains between 33 and 86 percent of the observed decline in robbery and motor vehicle theft in NYC between 1990 and 1999. Keizer, Lindenberg, and Steg (2008) tested the model in the Netherlands by selecting and monitoring several urban locations and arranging them in two di erent ways (‘orderly and disorderly’), at di erent times. They concluded that examples of disorder, like gra ti or littering, “can indeed encourage another, like stealing.� MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Box 3. Neighborhood Disorder and Crime Victimization in São Paulo As part of this report, we relied on a victimization survey done in São Paulo to assess the role that particular community characteristics, neighborhood disorder in particular, have on probability of criminal victimization from occurring. Research around commu- nities and crime has shown that higher levels of neighborhood disorder or broken win- dows theory may actually encourage criminal activities as they signal o enders that there are lower levels of guardianship in the neighborhood (Kelling and Cole, 1998). Data, Methodology, and Results. The survey had 1,986 respondents in 9 regions of São Paulo in 95 districts and 220 census tracts. Using multi-level model analysis, we examine the potential association between social cohesion and neighborhood disorder and a set of covariates at the individual and neighborhood levels on the likelihood of criminal victimization. The victim measure captures information on whether respon- dents have been victims in either one of the crimes explored in the survey (1 if the respondent has been a victim of crime in the past 12 months and 0 otherwise). The most salient �ndings of this preliminary analysis suggest that: š Neighborhood disorder appears to be statistically signi�cant and strongly asso- ciated with higher levels of victimization in the researched neighborhoods after controlling for other individual and neighborhood characteristics. š Social Disorder, a measure that captures the degree of disturbance in the neigh- borhood--for example, loud music or loud noises late at night or hearing gun shots during the night hours—is not associated with increased odds of being a victim of crime in São Paulo. š Social cohesion doesn’t seem to be associated with lower levels of victimization in São Paulo. These results are consonant with previous studies on social cohesion and crime victimization in Brazil where the levels of social cohesion in a neighbor- hood are not statistically signi�cantly associated with a lower probability of crime victimization. Conclusion. We tested the e ect that social cohesion and neighborhood disorder have on criminal victimization using the results of a victimization survey conducted in São Paulo in 2008. Using multi-level modeling, our results suggest that there is a strong and statistically association between higher levels of neighborhood disorder and greater odds of being victimized. CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? TARGETING HOTSPOTS AND POPULATIONS AT RISK THROUGH MULTISECTORAL STRATEGIES The spatial and demographic concentration of violent crime has led policy makers to de- sign crime-prevention approaches that address the multiple factors placing these terri- tories and populations at risk. One of the empirical regularities of homicides is their spatial clustering in what are known as “hotspots� of crime and violence. These hotspots tend to su er from multiple forms of urban risk or deprivation (high levels of poverty, lack of public services, high exposure to natural disasters, urban decay, etc.). Another empirical regularity, as we saw in Chapter 1, is for homicides to be disproportionally concentrated among poor male youth, most of whom also su er from multiple disadvantages at the individual level (e.g., malnutrition, low education), family level (lack of parental monitoring), relationship level (friends are gang members), and community level (living in violence hotspots). This accumu- lation of disadvantages and risk factors either at the territorial or demographic level is what the literature has identi�ed as the main driver of violent crime. Decades of research evidence, increasingly supported by new computer systems and soft- ware, has helped to con�rm this uneven distribution of crime according to physical and social characteristics of particular areas. As a result, a great body of analyses has also focused on evaluating policy responses targeted at these speci�c areas. For example, in an evaluation of nine cases of hotspot policing, including �ve large cities in the United States and one suburb in Australia, Braga (2007) found signi�cant levels of crime reduction in seven of the selected studies. The most successful of these initiatives are usually complemented with social inter- ventions targeted at di erent risks faced by the most vulnerable populations, such as youth. One approach to deal with multiple risks is to have multiple interventions that address risks one at a time. For instance, one way to deal with the risk of being out of school and idle in high violence communities is to have policies targeting children in these communities to keep them in school or providing incentives to return to school. For instance, small mon- etary transfers to families provided through conditional cash transfer programs targeted to the poor have proven to be an e ective way to encourage disadvantaged young people to complete secondary school in both Colombia and Mexico (Cunningham et. al. 2008). Chapter 4 presents evidence on the e ectiveness of CCT Bolsa Família in having this kind of impact in São Paulo. In general, all three cities studied in this chapter had several examples of single programs dealing with single risks. For instance, the Family School (Escola da Família), rolled out in São Paulo in 2003, aimed to reduce violence in schools by involving the community in school activities and the lives of students. The program opened 5,306 public schools on the weekends in 645 di erent municipalities. A study found a 36 percent reduction in criminal acts within the school or its surroundings between 2002 and 2005 (Waisel�sz and Athias 2005). MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Another approach is to have single interventions addressing multiple risks at the same time. This can be done either by making sure that multiple programs give priority to speci�c territory or population target or by designing a program with multiple components dealing with di erent risks. Each of the three cities has had a high pro�le example of this type of approach: Jardim Angela in São Paulo, Fica Vivo in Belo Horizonte, and Unidades de Policia Paci�cadora in Rio de Janeiro. Targeting hotspots through comprehensive local prevention strategies in Jardim Ângela, São Paulo. In the mid-1990s, Jardim Ângela was a hotspot, a São Paulo neighborhood de- clared one of the most dangerous places on the planet by the United Nations in 1996. As one of the most underprivileged neighborhoods in the city, Jardim Ângela was the �rst area to bene�t from an initiative launched by the São Paulo Municipal Government in 2000 to direct social programs toward the most violent districts in the city. This initiative included programs such as Renda Mínima, Bolsa Trabalho and Começar de Novo, designed to assist the most vulnerable families, to help youth �nd jobs, and to support the reentry of people over age 40 into the workplace. This initiative was accompanied by a series of community and youth programs focusing on conflict resolution, gun and alcohol control, and partnerships with the police. These programs are also believed to have contributed to the early onset of the decline in violence that is unique to São Paulo. All these e orts were guided by a local violence prevention program designed and implemented by a broad-based movement that included community and civic associations, nongovernmental organizations, police agencies, local government o cials and others. Between 2001 and 2007, homicides fell 76 percent in Jardim Ângela (Mizne 2008). Unfortunately, the lack of impact evaluations makes it impos- sible to know the extent to which these policies were responsible for the decline. Similar comprehensive local prevention interventions in other cities and neighborhoods that have been evaluated show signi�cant impact in crime reduction. This is the case of the Safer Cities Programme (SCP) implemented in England in the late 1980s in an attempt to reduce crime and fear of crime, with the ultimate goal of creating safer environments for community economic and social development. Although the program was part of a national initiative, its implementation was carried out at the local level. Among the di erent impact evaluations of the program, Ekblom et. al. (1996) found an overall reduction of 21 percent in burglary risks. Sequencing of these multiple interventions seems critical in contexts where social inter- ventions, whether carried out by the state or by NGOs and the private sector, are hindered by the overt control of the territory by drug tra ckers. It is very di cult to provide op- portunities in contexts where local governance is in the hands of gangs or organized crime, as in the case of Rio’s favelas. A certain level of peace seems to be a precondition for social development, and social development is a core ingredient for the sustainability of the peace process. As a result, good performing policies have usually combined three stages: �rst, breaking the cycle of violence by retaking control of the territory from the hands of violent CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? groups, second, placing substantial community or proximity police forces in those territories to help maintain that level of peace, and third, ensuring the provision and improvement of social services delivery, as well as additional social interventions targeted at speci�c at risk groups. Two examples of this are Fica Vivo in Minas Gerais and UPPs in Rio de Janeiro.10 In other contexts, where this territorial control is not so present, sequencing may be less of an issue, and control and prevention actions can be implemented in parallel. This was the case of successful hotspot interventions such as the already mentioned Operation Cease�re in Boston, and several other US cities, which focused on gun violence (illicit drug tra cking) and gang violence. Under these programs, stricter law enforcement and patrol in targeted areas was accompanied by extensive work with youth at risk and gang members, aimed at changing criminal behavior. Operation Peacekeeper in Stockton, California, trained youth outreach workers in community organizing, mentoring, mediation, conflict resolution, and case management, to use them as mentors and positive role models for youth in their respective communities. A pre/post comparative analysis showed that the program was as- sociated with an overall 42 percent decrease in the monthly number of gun homicide inci- dents in the city during the period in which it was implemented (1997-2002)(Braga 2008). Fica Vivo in Minas Gerais: combining hotspot and youth targeting through a sequenced strategy. The principal objective of the Fica Vivo program in Minas Gerais is to reduce ho- micides, particularly among youth, in urban areas with the greatest vulnerability to violent crime. It combines police interventions with social programs, carried out in cooperation with multiple stakeholders (military and civilian police, di erent municipal departments, NGOs, and the community itself). Minas Gerais’s Fica Vivo program, in addition to carefully targeting at-risk youth in high-crime neighborhoods, is also noteworthy for the sequencing of its crime control activities. Fica Vivo’s police intervention has two phases. First, a strategic police force aims to capture hardened criminals living in the neighborhood and to remove guns from the community. This phase is followed by the introduction of a community police force, the Special Police Group for High-Risk Areas (Grupo Especial de Policiamento em �reas de Risco, GEPAR), which seeks continuous interaction and collaboration with the community. A local community forum is organized monthly to discuss crime-related problems and coordinates strategies with the GEPAR police. For young people, the program o ers social support, as well as educational, recreational, and sports components, including workshops on violence prevention. The Fica Vivo program was inspired by the CeaseFire Project implemented in Chicago since the 1990s and was piloted in Belo Horizonte in 2002. In the following year the 10 Medellín, which during the 1980s and early 1990s became the most violent city in the world as drug cartels and later guerrilla militias took over control of the city’s marginalized hillside communities, followed a simi- lar approach to the one being implemented by the UPPs in Rio. The unprecedented peace period witnessed in the Colombian city in 2003-2008 was largely a result of increased preventive presence through the in- creasing establishment of police stations and patrols, coupled with increased municipal investments in local public works projects (WOLA 2011). MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME program was institutionalized by the Minas Gerais State Government, becoming the respon- sibility of the SEDS. Box 4. Impact Evaluation of Fica Vivo An evaluation of the pilot program showed a reduction of 34 homicides per 100,000 inhabitants every six months in the neighborhood where it was piloted, Morro das Pedras, leading to a 40 percent decline just in the �rst 12 months of the project (Bea- to Filho 2005). An external impact assessment of the program showed that Fica Vivo reduced homicides, although the e ect of the program was not homogeneous and increased with time. The evaluation used a di erence-in-di erences with propensity score-matching methodology (Azevedo, Peixoto and Andrade 2009). A more recent study revealed a 50 percent decline in homicide in the areas of the city where the program was implemented (SEDS 2010). A cost-bene�t analysis of its impacts also showed that the program had a 99 percent rate of return for society (Peixoto 2008). Unidades de Polícia Paci�cadora in Rio de Janeiro. UPP is a police peacekeeping initiative, launched in the City of Rio de Janeiro in December 2008, which aims to recapture territory until recently controlled by Rio’s drug-tra cking gangs. The UPP implementation has three phases. It starts with the initial tactical invasion of the community, usually led by the Special Police Operations Battalion (Batalhão de Operações Policiais Especiais, BOPE). It is then fol- lowed by the establishment of a permanent community policing force whose members are selected speci�cally for UPP service from the Military Police Academy and receive special- ized training in human rights, conflict prevention and mediation, and community policing. Finally, the public security aspects of the UPP program are accompanied by an integrated and targeted social program called UPP Social, administered by the Rio de Janeiro Municipal Government. The UPP Social consists of a coordination mechanism that maps government, private sec- tor, and civil society programs/interventions and channels them towards the achievement of speci�c goals aimed at consolidating peace and promoting long term social development of the favelas. These goals are operationalized through basic needs assessments and over- seen locally by management teams based in each favela. The UPP Social is launched in each new community through a community forum, designed to provide a forum for debate, open among government representatives, service providers, local businessmen and entre- preneurs, community leaders and residents. Together, UPP and UPP Social seek to provide sustainable public security and full democratic participation to residents of UPP-receiving communities. UPP and UPP Social promises to be a game changer in the �ght of Rio de Ja- neiro against drug tra c related crime and violence (Box 5). CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? Box 5. Extending the Right to live in Peace to Favela Residents by Changing the Rules of the Game for Rio de Janeiro’s Drug Traffickers: Unidades de Polícia Pacificadora (UPP) A long standing and seemingly intractable problem. Conflict between and among Rio’s three main drug-tra cking factions and the Military and Civilian Police have driv- en Rio’s homicide rate for most of the past 30 years. Public security experts estimate that more than half of Rio de Janeiro’s homicides are directly related to the organized drugs trade, resulting either from police action in faction-controlled areas or from con- flict among and between faction members, often with innocent victims caught in the cross�re (Amnesty International 2007; G. Soares 2008). The launch in December of 2008 of the Police Paci�cation Unit (UPP) initiative in the capital aimed to seize back control over territory formerly dominated by the drug tra ckers, and to provide public security and extend public services to the residents of favelas long cut o from the rest of the city. The UPP tries to change the rules of the game for drug tra cking in Rio de Janeiro. UPP does not represent a “war on drugs� with the explicit aim of capturing or killing as many gang members as possible and abolishing drug tra cking. Rather, the program seeks to deny drug-tra cking gangs control over territory, and to assert the state’s monopoly of the legitimate use of force across the capital. The expectation is that this change in the rules of the game will compel Rio’s drug-tra cking gangs to adjust their business model to conform more closely to the clandestine and less violent model em- ployed by their counterparts in Europe and the United States, without provoking an all- out war between organized crime and the state, such as that now playing out in Mexico. The program is expanding rapidly. As of December 2012[update], the program had deployed about 3500 in 28 UPPs throughout the city, covering more than 100 com- munities and reaching approximately 400,000 people. By 2014, the state government aims to deploy about 12,500 policemen in 45 favelas covering about 800,000 people or roughly 75 percent of favela residents in the city of Rio de Janeiro. Preliminary results are very encouraging. In areas served by UPP violent crime has fallen dramatically, while property value has increased. A recent analysis led by profes- sor Ignacio Cano, from the state university of Rio de Janeiro (UERJ), shows that violent deaths (e.g. homicides and deaths in encounters with the police) have decreased 78 percent in UPP areas from January 2006 to June 2011, while other non-violent crimes have substantially increased, such as threats (from 29.4 to 99 per 100,000 inhabitants), domestic violence (from 27 to 84.6 per 100,000) and rape (from 1.3 to 4.8 per 100,000) over the same period (UERJ and FBSP 2012).11 The study argues that these types of 11 Universidade Estadual do Rio de Janeiro (UERJ) and Fórum Brasileiro de Segurança Pública (FBSP) (Coord. Ignacio Cano). 2012. “Os Donos do Morro: Uma avaliação exploratória do impacto das Unidades de Polícia MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME crime could have increased either because now there is no “parallel� power of drug lords brutally enforcing the order upon favela residents and their relationships with neighbors and family members, or that crime reporting has increased with the arrival of the UPPs. Below are some earlier statistics for crime and police activity in Cidade de Deus, which according to police reports mirror results across paci�ed favelas. Preliminary results from public polls and ongoing research projects suggest the powerful e ects that the paci�cation process is having in the lives of those communities. A survey of favela residents found that 86 percent of those interviewed in favelas with UPPs believed that safety had greatly improved and 72 percent felt that outsiders had greater respect for them after the UPPs were installed. Among those without UPP, 72 percent considered UPP a positive initiative and 70 percent would support receiving the program (IBPS, 2010). Residents report feeling a great sense of relief and are enjoying re-claiming their public spaces and the “liberty to come and go� as they please at any time of day or night. Some bene�ts have already started to spill over to the neighbours of the favelas. A study by the Housing Union (SECOVI) found that houses located near UPP favelas in the South Zone had increased their rental value by 149 percent and sale value by 59 percent after the arrival of the UPPs (O Globo, 3/12/2010). Citizens from the favela and the “asfalto� regard it as a major success in the �ght against organized crime and is the main contributing factor to Governor Cabral Filho’s popularity. Cidade de Deus - Crimes Cidade de Deus - Atividade Policial 364 160 -71% 140 141 300 120 -49% 201 100 200 185 -84% 83 80 68 60 -82% 42 41 100 40 34 -88% 56 54 -83% -100% 47 20 17 11 20 6 6 8 2 1 2 0 0 0 Apreensão Prisões Ocorrências Auto de Homicídio Encontro de Roubo Estab. Roubo de Roubo de Roubo a Roubo em de Drogas em Plagrantes Resistência Doloso Cadáver Comerrcial Veículos Carga Transeuntes Coletivo Nov/2007 à Nov/2008 Nov/2008 à Nov/2009 Fonte - ISP - Instituto de Segurança Pública (nov 07-08 e nov 08-09) Paci�cadoras (UPPs) do Rio de Janeiro�. Rio de Janeiro: UERJ e Fórum Brasileiro de Segurança Pública. CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? INTEGRATED AND MULTI-STAKEHOLDER SUBNATIONAL CITIZEN SECURITY STRATEGIES Subnational governments’ role in developing integrated and multi-stakeholder citizen security strategies is often mentioned as an important contributor to reducing homicide, particularly in Minas Gerais and in some of São Paulo’s most violent municipalities. These strategies bring together the three elements of policy mentioned before (results oriented policing, risk factors, and targeted interventions). In addition, they are key to mobilizing mul- tiple stakeholders to participate in crime control and prevention activities. Cities and towns that participated in the Metropolitan Forum for Public Security—an initiative of the São Paulo Institute Against Violence (Instituto São Paulo Contra a Violência, ISPCV)—appeared to reduce their homicide rates by 6.3 percent per year between 1999 and 2003 (Waisel�sz and Athias 2005). The establishment of the Municipal Secretariat for Urban Security (2002) and the de- centralization of the São Paulo Government allowed local governments to have more control over urbanization policies, such as the installation of street lighting and other key services (Kahn 2005; Sou da Paz 2008). Assessing the impact of this type of policy is very di cult given the multiple programs and interventions and changes that are part of them and take place at the same time. One of the best known examples of this type of approach is the mu- nicipality of Diadema in São Paulo (Box 6). Box 6. Municipal-level strategies for crime prevention in Diadema In 1999, Diadema was the most violent municipality in São Paulo. Since then it man- aged to drop its homicide rate by 81.5 percent in eight years, from 111.62 per 100,000 in 1999 to 20.67 in 2007. Crime reduction was led by two municipal citizen security plans that encompassed a range of interventions including the three types of policies mentioned in this chapter (i.e. result oriented police reform, guns/alcohol control, hot spot and population at risk targeted programs). These plans and their preventive public policies were carried out through strong multi-stakeholder partnerships that included local, state and federal authorities, as well as civil society, religious and police groups. In 2001 the municipality developed its �rst citizen security plan focused on investing in a new technology system for the police and in prevention programs. This includ- ed the incorporation of MapInfo in 2002, which allowed for the targeting of criminal hotspots that could be crossed with vulnerability and social inequality maps, and the signing of an agreement with São Paulo state Government to access InfoCrim. A few highlights include: š Access to the newly gathered data allowed the identi�cation of youth as the group most a ected by crime and violence, as well as of a high number of vulnerable youth with learning disorders who had left school. Based on that, a project was MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME implemented to build life skills and strengthen citizenship among youth and pre- pare them to work. š Data also showed that 60 percent of the homicides in the city in 2001 had occurred between 11 p.m. and 6 a.m. and were located close to establishments that sold alcohol for immediate consumption. The response to that was the establishment of the successful ‘Bar Closure Law’, which stipulated an 11 p.m. curfew for all bars in the municipality. Estimates show that the law saved 11 lives each month over the following two years. š Other key interventions included public education on preventing crime and vio- lence (including campaigns against the abuse of alcohol and �rearms); community policing; and improvements in public security through environmental design (e.g. increased lighting, revitalization of public spaces, etc.) there was a 55% reduction in sexual violence against women, and an 80% reduction in alcohol-related prob- lems received at 24-hour health centers. In 2005, a second municipal security plan - “Diadema em Paz� – was developed, using the information collected in the previous years. The design and implementation of the plan followed a strong participatory approach that included public meetings with more than 800 community leaders and a traveling forum that brought together 1,500 people to discuss security issues during six months of meetings. This second plan focused on management, urbanization, criminality and risk factors, 17 security commitments involving the public authority, organized civil society, Military and Civil Police, and the municipal civil guard. It included the training of security forces in the use of the new technologies, the establishment of mediation centers to promote nonviolent conflict resolution, a Disque Denúcia hotline that allowed people to inform the police if they witnessed the illegal sale of alcohol, and an integrated video-monitoring center. Source: Miki 2008. Bogotá represents another case of integrated and multi-stakeholder initiatives that are associated with signi�cant crime reduction. A decade of multi-sector interventions in the city are linked to a decline in homicide rates from a peak of 80 per 100,000 inhabitants, in 1993, to 22 in 2004 (Cunningham et al. 2008). Strategies implemented included targeting risk factors such as gun and alcohol consumption control; capacity building of the police; initia- tives that promoted to cultural changes of citizens’ behavior; regeneration of urban spaces; and a set of institutional reforms. Another distinctive feature of this case was the establish- ment of an outside evaluation system through the program “How’s Bogota Going� (“Bogotá como Vamos�), by which three independent organizations - the Bogota Chamber of Com- merce, the newspaper El Tiempo, and a well-known local NGO – became responsible for fol- lowing closely the implementation of the local Security Plan, while also monitoring citizens’ CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? perceptions and indices of victimization. Bogota strategy also included a strong citizen par- ticipation component by providing training for community leaders on coexistence and citizen security at the community level. Although several studies have tried to measure the impact of several of these speci�c interventions, as in the case of Diadema it has also been di cult to know exactly what was the contribution of each of them in the overall positive results on crime reduction. Yet Bogota also suggests that there was an adequate combination of strate- gies (Cunningham et al. 2008). Institutional innovation to develop State level Public Security Strategy in Minas Gerais. The Minas Gerais State Government administration that took power in 2003 merged the Secretariats of Justice and Public Safety to create the new Secretariat of Social Defense (Secretaria Estadual de Defesa Social, SEDS). With the creation of the SEDS, for the �rst time in Brazil’s history, all institutions and agencies in charge of public security in the state were integrated under one Secretariat. These include: civilian and military police organizations, penitentiary administration, the public defender’s o ce, �remen, and agencies responsible for juvenile delinquents. Following the same results-oriented reforms that were applied in São Paulo, the SEDS de�ned a single strategy for public security in the state that included not only crime control but also proactive social and urban crime prevention policies. The strategy was part of a broader results-oriented reform process that the State Government introduced in several sectors; it involved de�ning goals and monitoring and evaluating the performance of state agencies in achieving those goals. SEDS’s strategy was organized around three pil- lars: integrated management of public safety (through the implementation of IGESP), prison system reform, and social prevention of crime. The prison system reform is also considered in the literature as one of the key factors that led to crime reduction in Belo Horizonte. It included the uni�cation of the state pris- on policy, which was previously segmented into di erent secretariats (Sapori 2007, Rocha 2010). The new Management Model for the Prison System (Modelo Referencial de Gestão do Sistema Prisional) placed the management of the prisons under SEDS, thereby allowing the Civil Police to start focusing on investigations, crime prevention and repression, and on the rehabilitation of criminals. Prior to this change, the Civil Police were in charge of the prison- ers in custody. Between 2003 and 2009, 44 prisons that used to be managed by them became SEDS’s responsibility. By the end of 2009, all prisons in Belo Horizonte and its metropolitan area had been either transferred to SEDS or deactivated (Ribeiro et. al 2004, Sapori 2007, Minas Gerais 2009). In addition, education, health, professionalization and legal centers were also implemented or augmented in most prisons as part of the reform. The social preven- tion approach of the crime pillar included the development of crime prevention programs carried out in partnerships with other public institutions (Civil and Military Police, judiciary, etc.), other government actors (state, municipal government) and civil society organizations (Beato 2007, Andrade 2006, Sapori 2007, Sapori and Andrade 2008, Batitucci 2010, Cruz 2005, Sousa 2009, Paula 2009). MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Most of these programs embraced a multidimensional approach to crime and violence reduction that placed a premium on partnerships between law enforcement institutions such as the police and judiciary, and between government actors at the state and municipal levels while also including civil society organizations (Beato 2007; Sapori and Andrade 2008; Batitucci 2010). The decentralized nature of several of the results-oriented reforms implemented in Minas Gerais and São Paulo brings into sharp relief the importance of front-line workers during program implementation. What perceptions do these frontline workers have about the drivers of crime reduction in their area, and about the relative e cacy of the programs being implemented there? How accurate are these perceptions, and how can these percep- tions a ect program implementation? A background paper was prepared for this report to address these questions using a positive deviant methodology (main results are summa- rized in Box 7). Box 7. Exploring the Perceptions of Frontline Workers about the Drivers of Crime Reduction – A Positive Deviant Analysis in Belo Horizonte The paper by Azevedo et al (2011) presents the �ndings of a positive deviant study designed to explore in a systematic manner the perceptions of those at the frontline about the factors correlated with the reduction of violent crimes in the city of Belo Horizonte. The study was based on a sample of 100 census tracts, which had similar trends of homicide between 2000-2003, but then observed opposite trajectories be- tween 2004-2009. Almost 500 frontline workers from the local Police and NGOs work- ing in those census tracts were interviewed face-to-face. Key �ndings include: š The perceptions of the frontline actors seemed accurate, as respondents from higher and increasing homicide areas were more likely to claim that the criminal activity in his/her census tracts seems worse when compared to either the AISP or the city of Belo Horizonte. š According to the systematic perception of respondents, crime reduction seemed to be correlated with the presence of selected programs in the past �ve years. Some of these programs include Fica Vivo, GEPAR, community police, IGESP, Con- flict Mediation, mobile courts, Open schools during the weekends, BH Ciudadania, Municipal Police (Guardia Municipal) and Bolsa Escola/Bolsa Familia. š In addition, the perception of improvement in the quality of implementation of these programs in the course of the last �ve years is also signi�cantly di er- ent among groups for the three main programs, namely, Praca de Esporte, GEPAR, IGESP. Areas in which the perception of the implementation programs had wors- ened, were more likely to have observed an increase in crime. CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? š These results should not be interpreted as evidence regarding the e ective- ness of such programs, however one could see them as a possible argument to prioritize these interventions on a future agenda of more rigorous impact evalua- tions. Since not only the perception of their presence, but also the variation in the quality of their implementation seems to be correlated, according to this sample of decision makers, with the observed changes in crime reduction. Source: Azevedo et al (2011) APPLICABILITY TO OTHER STATES? Pernambuco’s recent response to criminality mirrors in several ways the strategies adopted several years earlier in São Paulo and Minas Gerais. In an attempt to confront the rising rates of violence striking the state until the mid-2000s, in 2007 the government launched a cross-cutting, integrated public security strategy called Pact for Life (Pacto Pela Vida). The program used a multisectoral approach, consisting of 138 actions for crime control and prevention involving police and institutional reforms (e.g., the development of sophisti- cated information systems to map crime, increased police training and results-oriented po- licing, hotspot policing, the integration of police forces, community policing and enforcement of gun-control legislation) and prevention initiatives (e.g., social programs targeted to youth at risk, investment in education, health, cultural activities, etc.). The program also developed a system of permanent monitoring and oversight that includes representatives from civil society and di erent sectors. During the �rst four years of implementation of the program, homicide rates declined by 26,9 percent (from 53.1 in 2007 to 38.8 in 2010), while in almost all other eight states in the Northeast region rates increased signi�cantly over the same period, with the exception of a small decline in Rio Grande do Norte. A study by the State’s Social Defense Secretariat released in early 201112 claims that the program was responsible for the nearly 14 percent decline in the violent crime rate in 2010, which represented 523 lives saved in the state when compared to 2009. The study also showed that in four years, Recife had a 34.4 percent decrease in the same index per 100,000 inhabitants. The Center for Research on Criminality, Violence and Public Security Policies (NEPS) at the Federal Univer- sity of Pernambuco is currently developing a qualitative and quantitative study to evaluate the program. The current administration in Brasília appears to be following the lessons from these states and cities. The Secretary of Public Security Regina Miki was the architect of the Di- adema experience, from her position as Municipal Secretary of Public Security for eight years. 12 Pacto pela Vida Vira Exemplo�, http://www.diariodepernambuco.com.br/2011/01/27/urbana1_0.asp. The methodology followed by this study was not reviewed for this report, so its claims should be taken as tentative. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME The Secretary launched a renovated National Citizen Security Program (Programa Nacional de Segurança Pública com Cidadania, PRONASCI) and the Federal Government is investing in information systems to better understand data on crime and public security and develop stronger impact assessments of potentially transformative programs such as the UPP and UPP Social in Rio de Janeiro and Pacto pela Vida in Pernambuco. PRONASCI has a dual focus on territories and populations at risk. It tries to transform these hotspots into “territories of peace� with two major PRONASCI initiatives: Mulheres da Paz (Women of Peace) and Protejo (Protect), which were speci�cally designed to identify and proactively assist young people living in the most dangerous neighborhoods. Their implementation has varied across states, but no thorough impact evaluation of these programs has been carried out for far. CHAPTER 3. THE SUCCESS OF THE SOUTHEASTERN STATES: A ROLE FOR POLICY? CHAPTER 4. REDUCING CRIME AMONG YOUTH: THE ROLE OF EDUCATION POLICY MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME This chapter weighs the evidence for the role of education as a protective factor against crime and violence, with particular focus on its causal impact on aggression, crime and vio- lence among at-risk male youth. In Chapter 2 we showed that large male youth cohorts and elevated high school dropout rates are strongly correlated with observed homicide trends at the national and sub national levels in Brazil for the period 1998-2008. This chapter goes beyond what is suggested by the estimates in chapter 2, exploring the causal impact on homicides of reducing the high school dropout rate and of projects targeting at risk youth through conditional cash transfers for schooling. The focus of this chapter is youth and juvenile delinquency, motivated by the overrepre- sentation of youth in crime and violence. Youth account for a large proportion of crimes: 15- to 19-year-olds account for 20 percent of arrests for violent o enses in the US (Levitt and Lochner 2004). In São Paulo, among those crimes for which the age of the suspected o ender is known, between 20 and 25 percent of robberies, thefts and motor vehicle crimes are committed by individuals under the age of 18. Not only are late adolescence and early adulthood the peak ages of criminal o enses, but they also coincide with a sensitive pe- riod in the development and formation of academic, socio-emotional and behavioral skills (Blumstein and Cohen 1987; Cook and Laub 2002; Farrington 2007). Indeed, both cognitive and non-cognitive forms of human capital are strongly predictive of criminal behavior (see Lochner and Moretti 2004; Lochner 2010). Results shown in chapter 2 show the importance of the size of male cohorts and its correlation with the homicide rate at the municipal level, where larger cohorts of males 15-19 in particular, appear to be positively correlated with larger homicide rates. Causal evidence is generally limited for Brazil because of data access issues and lack of empirical identi�cation. This chapter presents novel results from a background paper (Chioda, De Mello and Soares 2010, hereafter CMS). In particular, CMS studied the contem- poraneous impacts of schooling on crime and violence. The paper falls into a strand of the literature focused on measuring the impact of interventions that influence youth schooling, whether along the extensive or intensive margin. The extensive margin refers to investment in schooling and the impact on crime produced by changes in the number of youth enrolled in school. The intensive margin refers to how long a given student stays in school conditional on initial enrollment; here we seek to measure whether there is an e ect on crime of chang- ing the length in the school day. This section is organized as follows: The channels through which education and education policy may have consequences for the incidence of crime and violence are �rst reviewed. The speci�c policy interventions and environments in São Paulo are then discussed, as are the details of the data. Finally, preliminary results of the analyses are provided, along with brief discussions of the probable mechanisms. CHAPTER 4. REDUCING CRIME AMONG YOUTH: THE ROLE OF EDUCATION POLICY CHANNELS CMS’s paper investigates the impact of two education policies on crime and violence. The �rst, Brazil’s conditional cash transfer (CCT) program, operates on adolescents along the ex- tensive margin of education (whether or not to enroll in school) while the second, a lengthen- ing of the school day among younger São Paulo students, operates on the intensive margins (the hours spent in school conditional on enrollment). This study relates to two strands of the literature on crime. The �rst considers the ability of poverty reduction programs to influence criminal behavior; the second investigates the role of human capital investments in decreas- ing crime and violence. Although strong correlations relating poverty and education to violent and illegal activities are well documented, studies identifying causal links are more limited and are mainly re- stricted to UK and US policy evaluations. To the best of our knowledge, this paper is the �rst to consider the relationship between conditional cash transfers and criminality. The types of interventions studied here operate through mechanisms that fall into two broad headings: income and human capital e ects. Resources made available by poverty reduction programs may mechanically lower the incentive or need to engage in economically motivated crimes. For instance, under a relaxed budget constraint, newly a ordable goods decrease the incentive or necessity to obtain them illicitly. Indeed, there is considerable evi- dence that economic conditions a ect crime rates. For instance, small but signi�cant reduc- tions in property crimes are found in response to declines in unemployment rates (Freeman 1995; Donohue and Levitt 2001; Raphael and Winter-Ember 2001), while Machin and Meghir (2004) and Gould, Weinberg and Mustard (2002) document a negative relationship between crime rates and the earnings of low-wage and low-skilled workers. In turn, DeFranzo (1996, 1997), Zhang (1997) and Hannon and DeFranzo (1998) study the link between welfare pay- ments and crime directly, �nding a negative relationship for serious crimes as well as prop- erty crime. However, the most compelling evidence on the income gradient of crime comes from a housing voucher program in Chicago, which in fact transferred the equivalent of 50 percent of household income to bene�ciaries. Jacob and Ludwig (2010) report close to a 20 percent decline in both violent and overall arrests in response to voucher receipt, implying an income elasticity of -0.4. In sum, the economic motivation for crime does not appear to be merely theoretical. In addition, supplemental income from poverty reduction programs may alter a household’s routines in a manner that exposes it to less risk of victimization and/or opportunities for de- linquency, such as by a ording parents more time for supervision (Heller, Jacob and Ludwig 2010). On the other hand, greater resources are also associated with purchases of normal goods, including criminogenic goods such as alcohol or narcotics that can induce loss of in- hibition or control and increase aggression. For instance, Dobkin and Puller (2007) document that consumption of illicit drugs increases among welfare recipients at the beginning of the MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME month, when their checks arrive. While holding other inputs constant, the theoretical e ect of increasing income is therefore ambiguous. However, policies that supplement family income are also associated with changes in the composition of household inputs, including invest- ments in human capital. In the context of the conditional cash transfers considered here, this link is required by their conditionalities. Income Channels t More resources relax HH budget constranits and imply less incentive to engage in economically-motivated crimes (-) t HHs change routines in a manner that exposes them to less risk of victimization and/or opportunities for delinquency (-) t More resources imply more purchases of normal goods, including criminogenic goods: e.g., alcohol or narcotics (+) t Additional family income is associated with changes in the composition of HH consumption and investment, e.g., human capital. (-) Other Channels t Human Capital: education alters both risk aversion and patience, favoring forward- looking decisions and less risky behavior (-) t Social networks are formed in school, with the potential for pro-social peer effects, assuming that those youth who are enrolled are positively selected (-) t "UUFOEJOHTDIPPMNFDIBOJDBMMZHFOFSBUFTBOincapacitation effect: criminal opportunities are more limited in school than on the streets (-) t "UUFOEBODFJNQMJFTconcentration of youth in a common location and facilitates their interaction, which may lead to increased conflict and more group-based delinquency (+) Both the CCT programs and the school-day shift change in São Paulo directly impact hu- man capital investments. Each is expected to a ect the propensity to engage in criminal or violent enterprises by altering the relative rewards of work and crime through a number of channels (Becker 1968; Lochner 2010). First, education may a ect the cost-bene�t calcula- tions of criminal participation by altering preferences and discount factors, inducing greater patience and placing more weight on future welfare, thereby favoring more forward-looking decisions (Becker and Mulligan 1997). Similarly, education is related to changes in psychic returns to crime (Lochner 2010) by rendering students more risk averse, which makes par- ticipation in crimes associated with substantial uncertainty in rewards or punishment more costly. An additional bene�t of education lies in the social networks that are formed in school and the potential for pro-social peer e ects (Glaeser, Sacerdote and Scheinkman 1996), as- suming that those youth who are enrolled in school are positively selected, in the sense of having lesser inclination toward crime. With positive returns to education in the labor market, education also raises expected future wages and, by extension, the current opportunity cost of crime, thereby discouraging delinquency (Becker 1968). In the context of incompletely ra- tional behavior, psychologists and neuroscientists have emphasized the link between human capital accumulation and higher levels of executive functioning. This relates to self-regulation skills such as inhibitory controls, problem-solving strategies, memory and self-monitoring. Previous research has shown that a range of social and behavioral skills related to aggres- sion, self-control, moral reasoning and emotional coping are predictive of criminal and re- lated behaviors, in some cases even in early childhood (Agnew 1992; Heckman, Stixrud and CHAPTER 4. REDUCING CRIME AMONG YOUTH: THE ROLE OF EDUCATION POLICY Urzua 2006; Hill et al., forthcoming). Finally, enrollment in school a ects criminal activity through two additional mechanisms. First, attending school is mechanically associated with an incapacitation e ect, according to which criminal opportunities are more limited in school than on the streets. Incapacitation has been established as an important channel through which education a ects contemporaneous criminality. For instance, Anderson (2010) docu- ments declines in juvenile crime rates at the time of increases in the minimum dropout age across US states, while Berthelon and Kruger (forthcoming) report concomitant declines in the wake of Chile’s length-of-school day reforms which, since 1997, have gradually extended the amount of time that students spend in school from 32 to 39 hours per week. However, a second, countervailing mechanism is associated with schooling: school attendance implies concentration of youth in a common location and facilitates their interactions, which may lead to increased friction among them and result in more conflict and more group-based delinquency. By exploiting exogenous variation in teacher in-service days, Jacob and Lefgren (2003) report that property crimes committed by juveniles decreases by 14 percent on days when school is in session (i.e., due to incapacitation), but that the level of violent crime con- temporaneously increases by 28 percent (as a result of concentration). In turn, Luallen (2006) estimates the causal e ect of school attendance on juvenile crime by exploiting teacher strikes as an exogenous source of school closings and likewise reveals declines in property crimes but increases in violent o enses. The net impact of education on contemporaneous criminality therefore depends on the relative importance of the above mechanisms, hence the need for an empirical evaluation. EVIDENCE FROM SÃO PAULO CMS’s paper investigates the impact of two education policies on crime and violence. Making use of a unique dataset combining detailed school characteristics with time-stamped and geo-referenced crime information from the City of São Paulo, it measures the contempo- raneous e ects on crime of interventions that directly influence youth schooling, both along the extensive and intensive margins. In order to identify these causal e ects, the authors use two di erent policies. They �rst consider the impact of increased school enrollment induced by the conditional- ity of the federal and municipal cash transfer programs, Bolsa Família and Renda Mínima. These programs disburse cash to low-income families provided that children in the house- hold meet minimal levels of school attendance. They exploit this conditionality (and the 2008 expansion of Bolsa Família) to quantify the e ect of a manipulation in education along the extensive margin on the incidence of an array of crimes. Although this poverty reduction pro- gram may a ect criminality through peer, incapacitation and human capital e ects, a fourth mechanism could also be at play in this case: the income e ect of additional resources in recipient households. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Box 8. Policy Details (1): Conditional Cash Transfers During the 2006-2009 period, two conditional cash transfer programs were operating in the city of São Paulo. These were the federal program Bolsa Família and the munici- pal program Renda Mínima. Bolsa Família is a federal minimum family income program created in 2003 that has been the subject of numerous studies. The program has di erent bene�ts for families at di erent income levels and with various conditionalities. The Basic Bene�t is a pay- ment of R$60 for families with monthly per capita income below R$70. In addition, there is the Variable Bene�t, according to which families with monthly per capita in- come below R$140 receive R$22 per child under age 15 (up to the maximum of three payments). Finally, starting in 2008, bene�ts were extended to adolescents, through the creation of the Variable Youth Bene�t. This bene�t paid R$33 per family member between the ages of 16 and 17 (up to a maximum of two payments), for families with monthly per capita income below R$140. The maximum bene�t value in Bolsa Família is R$200 per family, which applies to families with monthly per capita income below R$70, three children under age 15, and two adolescents between the ages of 16 and 17. Schooling and health condition- alities are also attached to the payments, although it has recently been claimed that they have not been seriously enforced. In terms of schooling, program participation requires school enrollment and 85% attendance for children between 6 and 15, and 75% attendance for adolescents between the ages of 16 and 17. São Paulo’s Renda Mínima, in turn, is a municipal minimum family income program that was created in 2006. In order to be eligible for the program, families must have lived in the City of São Paulo for at least two years, have a monthly per capita income below R$175, and have at least one child under age 16. The conditionalities are school enrollment and minimum attendance of 85% for children between the ages of 6 and 15, and ful�llment of a vaccination schedule for children under age 7. Renda Mínima’s bene�ts typically complement those of Bolsa Família. The maximum bene�ts allowed by Renda Mínima are (total value = Bolsa Família + Renda Mínima): R$140 for families with one child, R$170 for families with two children, and R$200 for families with three or more children. The second policy instrument considered involves a change in the length of the school day. Starting in 2007, the Secretariat of Education of the city of São Paulo began reducing the number of schools operating three-shift school days, in favor of the longer two-shift system. The new shift structure is intended to bene�t children through more time in school and also to make more rational use of school time (Prefeitura de São Paulo 2010). This reform enables the authors to estimate the impact of schooling on crime and violence along the intensive CHAPTER 4. REDUCING CRIME AMONG YOUTH: THE ROLE OF EDUCATION POLICY margin of schooling, conditional on enrolling in school, in a manner that is not confounded by any income e ects. As mentioned, the aim is to measure the short-run impact of education investments on crimi- nal behavior, in the form of the e ect of schooling (and income) that is likely driven by day-to- day changes in the desire and opportunity to commit crime. While important, this study does not investigate the longer term impacts of educational attainment on criminal participation that operate through changes in the returns to legitimate work, in the �nancial or psychic rewards of crime, or in preferences, for which there is already considerable evidence (see Lochner 1999; Lochner and Moretti 2001).13 Crime Data The paper uses data from several sources. First, crime data are from INFOCRIM, a COMP- STAT crime-tracking system of the Secretaria de Segurança Pública do Estado de São Paulo, the state-level law enforcement authority. It includes georeferenced police report data for all crimes committed in the city of São Paulo from 2006 through 2009. INFOCRIM includes all information available in the police report. Besides the latitude and longitude where a crime was committed, it includes the type of crime, time (or estimated time) of occurrence, and some characteristics of the perpetrator such as age and gender, when available. The analysis concentrates on theft, robbery, vandalism, violent crimes (homicides, attempted homicides and assaults), crimes against minors, and drug-related o enses. Records related to these types of occurrences encompass a total of 1,473,939 crimes over the four-year sample. School and Student Data School and student data cover the period between 2006 and 2009 and come from two sources. Data from the Secretariat of Education of the city of São Paulo on state and munici- pal schools identify the type of school, school location (longitude and latitude), and number of students covered by conditional cash transfer programs (Bolsa Família and/or Renda Mín- ima). For municipal schools, data are also available on the year in which the school switched from three to two shifts and average performance in standardized Mathematics and Portu- guese exams from 2007 to 2009 (students enrolled in municipal schools of São Paulo—2nd, 13 School-based programs focused on improving social development among at-risk children have shown posi- tive e ects in terms of criminal behavior through early adulthood. Empirical evidence suggests that school- age programs speci�cally targeting at-risk children/youth and designed to improve social development can yield positive long-term e ects on an array of outcomes ranging from educational attainment, conduct disor- der, and criminal behavior. (See Evaluation of Fast Track, the Montreal Longitudinal Experimental Study, and the Seattle Social Development Project.) MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME 4th, 6th, and 8th grades—take Mathematics and Portuguese exams every year, known as the Prova São Paulo). Most municipal schools cover elementary education up to 8th grade, while state schools typically cover 5th grade through high school. From the Brazilian School Census, additional data are gathered on the number and demo- graphic characteristics of teachers and students, and on school infrastructure. The charac- teristics are: gender, race and education of teachers; age, gender, race and current grade of students; and, for school infrastructure, number of classrooms, presence/absence of treated water, sanitation, a television, and number of computers available. Box 9. Policy Details (2): Change in School Shifts Historically, most municipal schools in the City of São Paulo operated on a three-shift school day. The �rst shift would run from 6:50 a.m. to 10:50 a.m., the second from 10:55 a.m. to 2:55 p.m., and the third from 3:00 p.m. to 7:00 p.m. Each group of children would therefore typically spend four hours at school, regardless of the shift to which they were assigned. Starting in 2007, the Secretariat of Education of the City of São Paulo decided to reduce the number of schools operating under three shifts and switch the entire municipal system to the two-shifts-per-day system. Under the two-shift system, children spend �ve hours in school, either from 7:00 a.m. to 12:00 p.m. or from 1:00 p.m. to 6:00 p.m. This is supposed to bene�t children through more time in school and also to make more rational use of school time, preparing the municipal system for a future transi- tion to school shifts of six or seven hours, depending on the grade (Prefeitura de São Paulo 2010). As a result of the change in policy, the percentage of municipal schools operating un- der the three-shift system declined from 67.1% in 2006 to 14.6% in 2009. This percent- age is likely to continue to decline over the coming years. Empirical strategy: Given the nature of the interventions and outcomes under consideration, the natural unit of analysis is the geographic area in the approximate vicinity of each school. Unfortunately, São Paulo does not have a strict geographic de�nition of school districts. Stu- dents in the city are assigned to the closest available school, so there is a high correlation between school location and place of residence. Arti�cial districts around each school are therefore constructed by assigning to a given school the area that is closer to it than to any other school. This is an important step that associates crimes in a given location to a particular school. Obviously, children do not circulate only in these areas and do not necessarily live there, but they are likely to live close to the school and spend part of their day there. If there is a CHAPTER 4. REDUCING CRIME AMONG YOUTH: THE ROLE OF EDUCATION POLICY high probability that children are near the school area for a signi�cant amount of time, this approach should be adequate. US evidence supports this strategy, since it reveals a concen- tration of crimes committed by youth in periods immediately following school hours, when children/adolescents are likely to remain around the school (Snyder and Sickmund 1999; Jacob and Lefgren 2003; and Gottfredson and Soulé 2005). The system of assigning students to schools in São Paulo also supports this approach, since children are expected to study in the closest school available. Thus, the measure of crime per school is the number of crimes that happened closer to a given school than to any other school (these crimes are “assigned� to that particular school). Therefore, the construction of the dependent variable depends on the schools included in the sample. To check the sensitivity of the results to the age group of children considered, municipal and state schools are considered separately. Municipal schools typically cover younger children, from elementary school up to eighth grade, while State schools cover older children, from �fth grade to the end of high school. Also, to account for the possibility of movement and social interactions across children from di erent schools, the presence of other schools (and students) in a certain neighborhood around a given school (a given kilometer radius) are explicitly controlled for.14 SUMMARY OF RESULTS The analysis concentrates on theft, robbery, vandalism, violent crimes (homicides, attempted homicides and assaults), crimes against minors, and drug-related o enses. Records related to these types of occurrences encompass a total of 1,473,939 crimes over the four years in the sample. In the analysis of the relationship between CCT-related schooling and crime, the correlation between a school’s enrollment of students receiving the Bolsa Família and surrounding crime is found to be indistinguishable from zero. As discussed above, however, raw correlations are di cult to interpret in this context, because several mechanisms mediate the relationship. In addition, this relationship is confounded by unobserved socioeconomic conditions that pre- vail in the areas surrounding schools, such as poverty and unemployment rates, which are likely to be correlated with both CCT receipt and crime rates. However, when CCT receipt is instrumented using Bolsa Família’s 2008 expansion to ado- lescents aged 16 to 17, large declines in all categories of crimes are uncovered. The e ects based on the full sample are similar in magnitude to those based on restricting at- tention to crimes committed exclusively on school days, suggesting that incapacitation does not exclusively account for the results. Instead, this evidence is consistent with income and 14 In the report and in the paper, details are provided on the identi�cation strategy and on the causal nature of the estimates. That discussion is omitted here for brevity. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME human capital e ects playing a role. In turn, the school shift change is found to have little impact on criminality, perhaps because the policy operates on youth who have already opted to enroll in school, and who are thus at a lower baseline risk of delinquency. The extensive margin (policy instrument: CCTs, conditionality): Among state schools which, as previously mentioned, typically provide high school instruc- tion, there is little relationship between the number of students receiving federal or state CCT and any of the various measures of crime. For instance, conditional on a number of school and neighborhood characteristics, a year-to-year increase of one student receiving a CCT in a given school is on average associated with a 0.006% increase in crime of any kind. However, this relationship is statistically indistinguishable from zero. A concern regarding estimating simple associations between the number of students receiv- ing either the Bolsa Família or Renda Mínima and crime is that these are plausibly positively correlated with unobserved neighborhood characteristics, which are themselves correlated with criminality and violence, including poverty rates or unemployment. To overcome this possibility, the 2008 expansion of Bolsa Família eligibility, which extended bene�ts to adoles- cents up to age 17, is exploited by constructing an instrument that is the interaction of the number of students ages 16 to 17 in 2006 and an indicator for years 2008 and 2009. All Crimes Robberies Thefts Violent Crimes Drug-Related Against Minors # of Bolsa Familia Students -0.391%*** -0.464%*** -0.118% -0.252%* -1.50%*** -0.782%** Note: Dependent variable is the # of crimes in each category that occurred in the neighborhood of a high-school. Reported estimates represent the percent decline in crime of each type resulting from an increase of one student receiving Bolsa Familia. Instrumenting for the number of youth receiving a CCT in this manner reveals the importance of unobserved characteristics, since the IV estimates di er considerably from their OLS and count data counterparts. In particular, they suggest that an additional student receiving a CCT at a given state school is causally related to (statistically signi�cant) 0.39, 0.46, 0.25, 1.50, and 0.78 percent declines in all crimes, robberies, violent crimes, drug-related of- fenses, and crimes against minors, respectively. Although the latter two categories ex- perience the largest declines as measured by their semi-elasticities to the receipt of Bolsa Familia, such o enses are relatively rare events (only between 3 and 5 incidents per year per school neighborhood) particularly compared to robberies, which occur at an average rate of 433 per year per school district, such that the largest decline in number of incidents are among robberies. CHAPTER 4. REDUCING CRIME AMONG YOUTH: THE ROLE OF EDUCATION POLICY The intensive margin (policy instrument: lengthening of the school day): The results suggest little relationship between school shift length and crime. To the ex- tent that the incapacitation and concentration e ects are non-zero, this evidence suggests that they roughly o set one another, such that any gains resulting from keeping youth in school longer are unraveled by greater conflict or crime in school. However, the null results on nonviolent crimes such as thefts, which are unlikely to occur in schools and are therefore not a margin along which one would expect to observe a concentration e ect, suggest that even the incapacitation e ect is weak (at least along this margin). All Crimes Robberies Thefts Violent Crimes Drug-Related Shift Change 39.38 6.95 25.65 4.61 0.95 Indicator [22.96]* [5.41] [15.38]* [3.67] [0.91] Notes: For ease of interpretation, OLS estimates are reported. However, estimating negative binomial count data models reveals no statistically significant impacts. An alternative explanation for the �nding that there is no relationship between the length of the school day and crime is that this manipulation along the intensive margin of education does not capture those youth who are at higher risk of criminality. In particular, the interven- tion a ects only youth who are already positively selected, i.e., those who have previously opted to enroll in school and who are at lesser risk of delinquency. In addition, the reform a ects only municipal schools, which typically serve younger children (only three percent of students were age 16 to 17 in 2007) CONCLUSIONS AND WAY FORWARD MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME The landscape of violent crime in Brazil has undergone a profound transformation over the last ten years. As this report shows, for the �rst time in decades Brazil’s homicide rate is lower than the average for Latin America. Cities that were synonymous with insecurity and crime, such as São Paulo and Rio de Janeiro, have been the main drivers of this decline, plac- ing them, especially São Paulo, at the same level of change as the much-heralded cases of New York City and Bogotá. Unfortunately, most states in the Northeast and many in the North have experienced a signi�cant deterioration of their levels of public safety. We built a unique dataset to analyze the drivers of the national level decline as well as the heterogeneity across regions. At the national level, the reduction of the male cohort of 15- to 19-year olds, the reduction in concentration of income, and the reduction in high school dropout rates appear to be highly correlated with the decline in homicide rates observed fol- lowing 2003. Why did the Southeastern states do so much better than the North and North- east? What explains variation across regions? Our results suggest that a higher reduction in income inequality, school dropout rates and better labor market conditions in the Southeast are consistent with the observed reduction in homicides. In the Northeast, however, we ob- serve a smaller reduction in inequality compared to all other regions, and an increase in high school dropout rates. In addition, the Northeast’s increase in urbanization is also associated with increases in crime. For the Northeast and Southeast regions, we �nd that changes in coe cients of these variables largely account for most of the observed change in homicides since 2003. This means that the marginal propensity to commit crime is very sensitive to changes in these variables. Crime prevention policies should therefore pay special attention to inequality, high school dropout rates, and urbanization as strong risk factors to be tackled. However, we should remember that these are simple correlations, which do not capture cau- sation. The policy recommendations stemming from these results should thus be cautious, since these variables could be simply markers of underlying causal mechanisms. Our analysis of the residuals suggests that the variation across regions could potentially be related to di erences in policies. What are the policies that have distinguished the Southeast from the rest? The most important appear to be: results oriented policing, gun and alcohol control, programs targeting youth at risk and hot-spots of crime and violence, all guided by integrated and multi-stakeholder citizen security subnational strategies. These policies are not o ered as prescriptions of what governments should do since the evidence basis for their e ectiveness is not strong enough to establish causal relationships; even when they are, their external validity or transferability is not established. Instead they are o ered as examples of policies meriting careful further study through more robust analytical work, impact evaluations, and controlled policy learning. In particular, this learning agenda should include �ve themes that appear to have been important in the policy experience of the south- eastern states: 1. Reduce youth fragility. Although there is little that policy can do to influence the type of demographic shift that contributed to the recent national crime decrease, policies CONCLUSIONS AND WAY FORWARD can be put in place to reduce the fragility of youth. This report has summarized the experience of programs which can o er useful insights on how to adopt a multi- sectoral, territorially based strategy focused on youth at risk. In addition, compelling evidence from the CCT evaluation shows the power of education as a key protec- tive factor, and the value that instruments such as CCTs could have in reducing the fragility of youth at risk. The study found that an additional student receiving a CCT at a given state school is causally related to 0.39, 0.46, 0.25, 1.50, and 0.78 percent declines in all crimes, robberies, violent crimes, drug-related o enses, and crimes against minors, respectively. 2. Develop comprehensive crime-prevention strategies that combine territorially based interventions targeting crime and violence hotspots with generalized policies that reduce risks across the board. Reviewing the determinants of violent crime identi- �ed in this study gives a sense of the multiple factors and dimensions influencing it: rapid urbanization, increased circulation of �rearms, and reduced inequality are all complex phenomena that call for comprehensive strategies dealing with di erent risks and determinants through di erent instruments. There is no single magic bullet to reduce violent crime. Some risk factors are of an eminently territorial nature, as con�rmed by the victimization survey of São Paulo, and as further underscored by the e ect of rapid urbanization as a determinant of national crime. Slum upgrading and urban renewal policies that transform decaying and risky environments into places where communities can thrive should be combined with broader social and economic policies that enable economic opportunities to emerge in these territories. 3. Sequence and integrate crime-control and crime-prevention activities. It is very dif- �cult to provide opportunities in contexts where local governance is in the hands of gangs or organized crime. As the recent example of UPP and UPP Social reveals, in contexts such as the slums of Rio de Janeiro, peace is a precondition for social devel- opment, and social development is a precondition for the sustainability of the peace process. The apparent synergies among UPP, UPP Social and social programs in Rio de Janeiro are an example of the gains to be made through sequenced and integrated approaches. The paci�cation e ort can allow social programs to reach out to bene�- ciaries previously excluded from the program. The UPP Social is actively seeking to facilitate access to high schools so that the newly enrolled children can meet the pro- gram’s conditionality. The synergies from a sequenced approach are evident in many of the experiences reviewed in this report. GEPAR and Fica Vivo in Minas Gerais and the experience of Jardim Ângela in São Paulo are other examples of this approach. 4. Introduce a results orientation that rewards innovation and performance account- ability. Both Minas Gerais and São Paulo were at the forefront of e orts to reform the police through the adoption of results- and information-based management, through better integration of the civilian and military police, and through improved training MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME and professionalization of the police force. Minas Gerais created a Secretaria de De- fesa Social to provide strategic leadership and coordination for this agenda and hold the di erent actors accountable for agreed targets. The success of policies such as IGESP is an example of the results achieved through this approach. Usually, these innovations were founded on strong partnerships with universities and/or with the private sector and civil society, which lead to the next point on the knowledge and information platform that should underpin these e orts. 5. Build monitoring and evaluation systems that enable shared diagnostics, policy learning and experimentation among the key actors involved in the public safety agenda. Several of the innovations presented in this report were the result of care- fully designed pilots, accompanied by impact assessments that guided decisions re- garding their rollout. Fica Vivo in Minas, UPP and UPP Social in Rio, Disque Denúncia in São Paulo are examples of this. Assisting government teams in developing the skill set required to perform this controlled “learning-by-doing� approach, and tools such as the positive deviant analysis piloted in Belo Horizonte as part of this study, can be part of the repertoire of tools to identify areas that require more rigorous impact as- sessment. The future research agenda should also focus on understanding the dynamics of violent crime in the states that have su ered rapid increases in violent crime, particularly in the Northeast and North regions. As annex V shows, this includes states like Bahia, Alagoas, Maranhão, Para, Rio Grande do Norte, Paraiba, and Amazonas. This study has hinted at some of the factors that might explain the worsening of security in these states and capital cities: rapid urbanization, the e ect of dropout rates, and increases in inequality. Further studies should try to unpack these factors and assess more carefully their relative contribution to insecurity. Looking ahead from a policy perspective, Brazil needs to consolidate and deepen the gains made in the Southeast and tackle head-on the challenge of the Northeast. Fortu- nately, there is good news coming from the Northeast as well. Pernambuco and its capital city, Recife, which for much of the past decade had the dubious distinction of being by far the country’s most violent major metropolitan area, have seen homicide rates decline since 2008. Perhaps not coincidentally, Pernambuco is the only state in the Northeast to have implemented substantial public security reforms based on experiences from the Southeast. Other Northeastern states, including Bahia, have pledged to follow suit. The Federal Gov- ernment is investing in information systems to better understand data on crime and public security and develop stronger impact assessments of potentially transformative programs. Brazilians are well known in the world for their life-loving attitude, their warmth and ca- pacity to enjoy human fellowship. It is only �tting, then, that the country has started to put the blemish of crime and violence behind it and to build a future in which the right to live, and to live in peace, becomes one more thing about which Brazilians can feel proud. CONCLUSIONS AND WAY FORWARD REFERENCES MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Agnew, Robert (1992). Foundation for a general strain theory of crime and delinquency. Crimi- nology, 30(1): 47–87. Alda, Erik (2011). Social Cohesion, Neighborhood Disorder and Criminal Victimization: Evi- dence from São Paulo. Mimeo. Amnesty International (2005). “Eles Entram Atirando: Policiamento de comunidades social- mente excluídas no Brasil. London: Amnesty International Press. Anderson, D. Mark (2010). In School and Out of Trouble? The Minimum Dropout Age and Ju- venile Crime. Mimeo. Andrade, S. C. P. de. (2006). Polícia Bipartida: uma reflexão sobre o Sistema Policial Mineiro. Dissertação de mestrado. Belo Horizonte, EG/FJP. Andrade, M. V.; Peixoto, B. (2008). “Avaliação Econômica de Programas de Prevenção e Con- trole da Criminalidade no Brasil�. In: Beato Filho, C. C. (Org.). Compreendendo e Avali- ando: Projetos De Segurança Pública. Belo Horizonte, UFMG. Alzir das Chaves, Roberto and Cesário de Sá, Antonio. Public Security in Rio de Janeiro. Pre- sentation at the Wilson Center for Scholar on March 16, 2011. Batitucci, E. C. A Emergência do Pro�ssionalismo na Polícia Militar de Minas Gerais (1969-2009). Tese. Doutorado Sociologia, UFMG. Belo Horizonte: FAFICH/UFMG, 2010. Beato Filho, Cláudio (2005). Crime and Violence Diagnostics and Information for Urban Safety Audits. Washington D.C.: World Bank. Becker, G.S., (1968). “Crime and Punishment: An Economic Approach�, Journal of Political Economy, 76 (2), pp. 169-217. Becker, Gary and Casey B. Mulligan (1997). The Endogenous Determination of Time Prefer- ence. Quarterly Journal of Economics, 112(3), 729–58. Biderman, C., Mello, J.M.P., and Schneider, A. (2010). Dry Law and Homicides: Evidence from the São Paulo Metropolitan Area. Economic Journal, v.120, 157–182. Braga, Antonio, Kennedy, David; Piehl, Anne; and Elin Waring (2000). “The Boston Gun Project: impact evaluation �ndings�. Cambridge: Harvard University. ______ (2008). “Pulling Levers Focused Deterrence Strategies and the Prevention of Gun Ho- micide�. Journal of Criminal Justice 36(4): 332-43. Brandon C. Welsh and David P. Farrington. (2007) Save Children From A Life Of Crime. Crimi- nology & Public Policy. Volume 6, Issue 4. Bratton, William J. and Smith, Dennis C. (2001) “Performance Management in New York City: Compstat an revolution in police management.�, in Dall. W. Forsythe, ed., Quicker Bet- ter Cheaper? Managing Performance in American Government. Albany, NY: Rockefell- er Institute Press. REFERENCES Briceño-León and Zubillaga (2002). Violence and Globalization in Latin America. Current So- ciology January 2002 vol. 50 no. 1 19-37. Buonanno, Paolo, 2003. The Socioeconomic Determinants of Crime. A Review of the Litera- ture. Working Papers 63, University of Milano-Bicocca, Department of Economics. Buonanno, Paolo and Leone Leonida, 2005. Non-linearity between Crime and Education: Evi- dence from Italian Regions, University of Milano-Bicocca, Department of Economics. Bushway, Shawn D. , Cook, Philip J. and Phillips, Matthew, The Net E ect of the Business Cycle on Crime and Violence (July 13, 2010). Duke Department of Economics Research Pa- per. Available at SSRN: http://ssrn.com/abstract=1655741 Calderón, Valentina and Laura Chioda (2011, unpublished draft). Trends, elasticity and hyster- esis of crime in violence in Brazil (2003-2008). Cerqueira, D., Mello J.M.P. Menos armas, menos crimes: o emblemático caso de São Pau- lo (2009). Available at: www.ence.ibge.gov.br/pos_graduacao/mestrado/seminarios/ res30.ppt. Accessed on March 9, 2011. Chioda, Laura, João Manoel De Mello and Rodrigo Soares (2010). Does Education Reduce Crime & Violence? Evidence from Bolsa Família and Extending the School Day in São Paulo. World Bank. Chiu, H. and P. Madden (1998), “Burglary and income inequality�, Journal of Public Economics 69, 123-41. Cook, Philip J. and Jens Ludwig (2002). Gun Violence: The Real Costs. NY: Oxford University Press. ______ and John H. Laub (2002). After the epidemic: Recent trends in youth violence in the United States. In Michael Tonry, (ed.), Crime and Justice: A Review of Research, Vol. 29. Chicago: University of Chicago Press. Corman, H. y Mocan N. (2002) “Carrots, Sticks and Broken Windows�, National Bureau of Eco- nomic Research, Working Paper No. 9061. CRUZ, M. V. G.; BATITUCCI, E. C. (2006). “Novos Meios de Articulação Institucional na Segu- rança Pública: estudo de caso em Belo Horizonte, Minas Gerais�. In: SLAKMON, C; MACHADO, M. R. e BOTTINI, P. C. (orgs.). Novas Direções na Governança da Justiça e da Segurança. Brasília/DF, Ministério da Justiça. Cunningham, Wendy, Linda McGinnis, Rodrigo Garcia Verdu, Cornelia Tesliuc, and Dorte Verner. (2008). Youth at Risk in Latin America and the Caribbean: Understanding the Causes, Realizing the Potential. Washington, DC: World Bank. de Mello, Joao M. (2009). Does Crime A ect Economic Decisions? An Empirical Investigation of Savings in a High-Crime Environment (with Eduardo Zilberman), The B.E. Journal of Economic Analysis & Policy. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME de Mello, Joao M. and Alexandre Schneider (2007). Mudança Demográ�ca e a Dinâmica dos Homicídios em São Paulo (with), Revista São Paulo em Perspectiva. DeFranzo, James (1996). Welfare and Burglary. Crime and Delinquency, 42, 223–229. DeFranzo, James (1997). Welfare and Homicide. Journal of Research in Crime and Delinquency, 34, 395–406. Dobkin, Carlos and Stephen Puller (2007). The E ects of Government Transfers on Month- ly Cycles in Drug Abuse, Hospitalization, and Mortality. Journal of Public Economics, 91(11–12), 2137–57. Donohue, John, and Steven Levitt (2001). The Impact of Legalized Abortion on Crime. Quarter- ly Journal of Economics, 116(2), 379–420. Dreyfus Pablo and Júlio César Purcena (2009). Pegando o touro pelos chifres. Os efeitos de medidas de controle na indústriia brasileira de armas pequeñas. Flacso Brasil junho Ekblom, Paul, Ho Law and Mike Sutton, with assistance from Paul Crisp and Richard Wiggins. (1996). Safer Cities and Domestic Bulgary. Home O ce Research Study 164, London, England: Home O ce. Ehrlich, I., (1996). “Crime, Punishment, and the Market for O enses�, Journal of Economic Perspectives, 10 (1), pp. 43-67. Fajnzylber, P., Lederman, D. and N. Loayza (2002). “What Causes Violent Crime?�, European Economic Review, 46 (2), pp.1323-1357. Freeman, R.B., (1999). “The Economics of Crime�, Handbook of Labor Economics, ch. 52, 3, pp. 3529-3571. ______ (1996). “Why Do So Many Young American Men Commit Crimes and What Might We Do About It?�, Journal of Economic Perspectives, 10 (1), pp. 25-42. ______ (1995). The Labor Market, in Crime, James Q. Wilson and Joan Petersilia, eds., San Francisco: ICS Press, pp. 171–192. ______ (1994). “Crime and the Job Market�, in Wilson, J. Q. and J. Petersilia (Eds.), Crime. San Francisco: ICS Press. ______ (1991). “Crime and the Employment of Disadvantaged Youth�, NBER Working Paper no. 3875. Glaeser, E.L. and B. Sacerdote (1999). “Why is there more crime in cities?�, Journal of Political Economy, 107 (6), pp. S225-S228. ______ and J.A. Scheinkman (1996). “Crime and Social Interactions�, Quarterly Journal of Economics, 111, pp. 507-548. REFERENCES Grogger, J., (1998), “Market Wages and Youth Crime�, Journal of Labor Economics 16, 756-791. ______ (1995). “The E ect of Arrest on the Employment and Earnings of Young Men�, Quar- terly Journal of Economics, 110 (1), pp. 51-72. ______ (1991), “Certainty vs. Severity of Punishment�, Economic Inquiry, 29, pp. 297-309. Gottfredson, Denise C. and David A. Soulé (2005). The Timing of Property Crime, Violent Crime, and Substance Use among Juveniles. Journal of Research in Crime and Delinquency, 42(110), 110–120. Gould, Eric D., Bruce A. Weinberg, and David B. Mustard (2002). Crime Rates and Local Labor Market Opportunities in the United States: 1979–1997. Review of Economics and Sta- tistics, 84(1), 45–61. Hannon, Lance, and James DeFranzo (1998). Welfare and Property Crime. Justice Quarterly, 15, 273–287. Hartung, Gabriel (2009). Ensaios em Demogra�a e Criminalidade. Rio de Janeiro: Fundação Getúlio Vargas. Heckman, James J., Jora Stixrud, and Sergio Urzua (2006). The e ects of cognitive and non- cognitive abilities on labor market outcomes and social behavior. Journal of Labor Economics, 24(3): 411–482. Heller, Sara B., Brian A. Jacob and Jens Ludwig (Forthcoming). Family Income, Neighborhood Poverty, and Crime. In P. Cook, J. Ludwig and J. McCrary (eds.), Controlling Crime: Strat- egies and Tradeo s, Chapter 11, Chicago: University of Chicago Press. Imrohoroglu, A., Merlo, A. and P. Rupert (2001). “What Accounts for the Decline in Crime?�, Federal Reserve Bank of Cleveland, wp 0008. Imrohoroglu, A., Merlo, A. and P. Rupert (2000). “On the political economy of income redistri- bution and crime�, International Economic Review, 41 (1) pp. 1-25. Instituto Brasileiro de Geogra�a e Estatísticas (IBGE). (2010). Pesquisa Nacional por Amostra de Domicílios. Características da Vitimização e Acesso à Justiça no Brasil. Rio de Ja- neiro: IBGE. Instituto de Estudos Socioeconômicos (INESC). (2010). Segurança Pública e Cidadania: Uma Análise Orçamentária do PRONASCI. Brasília: INESC. Jacob, Brian A. and Lars Lefgren (2003). Are Idle Hands the Devil’s Workshop? Incapacitation, Concentration, and Juvenile Crime. American Economic Review, 93(5), 1560–1577. ______ and Jens Ludwig (2010). The E ects of Family Resources on Children’s Outcomes. Working Paper, University of Michigan. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME Keizer, K., Lindemberg, S; Steg, L (2008) “The Spreading of Disorder�. Science 322 (5908): 1681-5. Kelling, George L. and Sousa Jr., William H. (2001). Do Police Matter? An Analysis of the Im- pact of New York City’s Police Reforms. NY: Manhatan Institute for Police Research. Khan, Tulio, and Rizzi, Renata (2010). On the relationship between crime and economic per- formance in Brazil after 1994 using time series econometric techniques. Unpublished paper. ______ and Zanetic, André (2005). O Papel dos Municípios na Segurança Pública. São Paulo: Secretaria Nacional de Segurança Pública. Kovandzic, Tomislav, Vieraitis, Lynne M. and Boots, Denise. Does the Death Penalty Save Lives? New Evidence from State Panel Data 1977-2006. (2006). Forthcoming in Crimi- nology &Public Policy. Kvalseth, Jarald O. (1997). A Note on the E ects of Population Density and Unemployment on Urban Crime. Criminology Journal, Criminology, Volume 15, Issue 1, pages 105–110, May 1977. Levitt, S.D. and L. Lochner (2001). “The Determinants of Juvenile Crime�, in Gruber, J. (Ed.), Risky Behavior Among Youth: An Economic Analysis. Chicago: The University of Chi- cago Press.Lima, R.S. (2010). Diagnóstico de la violencia y la criminalidad en las ciu- dades de São Paulo y Río de Janeiro. Serie Documentos Electrónicos N. 5, agosto 2010. Santiago de Chile: FLACSO-Chile Lochner, L., (1999). “Education, Work and Crime: Theory and Evidence�, Working Paper No. 465, Center for Economic Research (RCER) - University of Rochester. Lochner, L. and E. Moretti (2001). “The e ect of education on crime: evidence from prison inmates, arrests, and self-reports�, NBER Working Paper no. 8605. Lochner, Lance and Enrico Moretti (2004). The E ect of Education on Crime: Evidence from Prison Inmates, Arrests, and Self-Reports. American Economic Review, 94(1), 155–189. Lochner, Lance (2010). “Education Policy and Crime.� NBER Working Paper 15894. Lott, J.R. (2004). More Guns, Less Crime: Understanding Crime and Gun-Control Laws. Chi- cago: University of Chicago Press. Luallen, Jeremy (2006). School’s out... forever: A study of juvenile crime, at-risk youth and teacher strikes. Journal of Urban Economics, 59(1), 75–103. Machin, Stephen, and Costas Meghir (2004). Crime and Economic Incentives. The Journal of Human Resources, 39(4), 958–979. REFERENCES Miki, Regina. A experiência de Diadema em políticas públicas e a segurança cidadã (2008) In: VELOSO, Fernando e FERREIRA, Sérgio Guimarães (Orgs). É possível: gestão da segu- rança pública e redução da violência. Rio de Janeiro: Casa das Garças. Pan American Health Organization (PAHO). www.paho.org. Accessed on March 12, 2011. Prefeitura Municipal de São Paulo (2010). “Balanço da Educação aponta avanços na rede de ensino público.� (http://www.prefeitura.sp.gov.br/portal/a_cidade/noticias/index. php?p=34456, posted on January 5, 2010). Raphael, Steven, and Rudolf Winter-Ember (2001). Identifying the E ect of Unemployment on Crime. Journal of Law and Economics, 44(1), 259–283. Secretaria de Estado de Desenvolvimento Social (SEDS) de Minas Gerais, accessed in March 2011, https://www.seds.mg.gov.br/. Snyder, Howard N. and Melissa Sickmund (1999). Juvenile O enders and Victims: 1999 Na- tional Report. US Department of Justice Programs, O ce of Juvenile Justice and De- linquency Prevention, Washington DC. Soares, Sergei and Natália Sátyro (2009). “O Programa Bolsa Família: Desenho Institucional, Impactos e Possibilidades Futuras.� IPEA Working Paper n1424. Souza, M.F.M, Macinko, J., Pereira, A., et al. (2007). Reductions in Firearm-Related Mortality and Hospitalizations in Brazil After Gun Control. Health A airs. 26 (2). Soares, Rodrigo R. & Viveiros, Igor (2010). “Organization and Information in the Fight against Crime: An Evaluation of the Integration of Police Forces in the State of Minas Gerais, Brazil,� IZA Discussion Papers 5270, Institute for the Study of Labor (IZA). Veloso, Fernando and Sérgio Guimarães Ferreira (Orgs) (2008). É possível: gestão de segu- rança pública e redução da violência. Rio de Janeiro: Contra Capa. Waisel�sz, Julio Jacabo. (2011). Mapa da Violência 2011 – Os Jovens do Brasil. São Paulo: Instituto Sangari. ______ (2010). Mapa da Violência 2010. São Paulo: Instituto Sangari. ______ and Athias, Gabriela (2005). Mapa da Violência de São Paulo. Brasília: UNESCO. Wilson, J.Q. & Herrnstein, R. (1985). Crime and Human Nature. New York: Simon and Schuster. Washington O ce on Latin America - WOLA (2011). Tackling Urban Violence in Latin America: Reversing Exclusion trough smart policing and social investment. Washington, DC. World Bank (2006). Crime, Violence, and Economic Development in Brazil: Elements for Ef- fective Public Policy. Poverty Reduction and Economic Management Sector Unit: Latin America and the Caribbean Region. Washington, DC: World Bank. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME ______ (2005). Case Study: Reduction of Crime in Bogota: A Decade of Citizen’s Security Poli- cies.� Washington, DC: World Bank Working Paper Series 35128. Zhang, Junsen (1997). The E ect of Welfare Programs on Criminal Behavior: A Theoretical and Empirical Analysis. Economic Inquiry, XXXV(1),120–137. Zimring, Franklin E. (2007). The Great American Crime Decline. Oxford: Oxford University Press. ______ (2011 – forthcoming). The City that Became Safe: New York and the Future of Crime Control. February 3, 2001. Oxford University Press. REFERENCES ANNEX MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME ANNEX I. DATA SOURCES In Brazil there are three main sources of information that capture various forms of violence: Health and Victimization Surveys (HVS), Health Ministry’s Mortality Records and Statistics (DATASUS), and Police Administrative Records (RPM). This report used DATASUS informa- tion for the national level analysis on trends and determinants (chapters 1 and 2), and police data for the analysis of Sao Paulo, Minas Gerais, Rio de Janeiro (chapters 3-6). Victimization surveys were only used for the case of Sao Paulo. Overall DATASUS is more reliable than police records. However, in the case of the southeast, the state governments have developed accurate information systems that show very similar trends to the ones from DATASUS. There have been studies that estimate the discrepancies in homicide data in Brazil. For in- stance, Monteiro et. al. (2003) compare homicide data from DATASUS and RPM in Minas Gerais. There may be discrepancies in the homicide rates that increasingly vary as more dis- aggregated is the data. DATASUS and RPM have most consistency over time and across ag- gregation levels. Most studies, including this one, conclude that the records on both systems are accurate across states and among cities with more than 100,000 inhabitants. However, for cities below this population threshold, inconsistencies emerge mostly on the RPM data because under-reporting rises due to hospital deaths that are not followed up by police, but were caused by violence. National level analysis. Our focus of interest, dynamics and determinants of violence at the national level as well as at state and municipal levels, imposes a restriction on the type of data available. Only Health Ministry’s Mortality Records and Statistics (DATASUS), and Police Administrative Records (RPM) data allows having yearly data by municipality and state in Brazil. This report opted to use DATASUS as primary source of data to be complemented whenever needed by RPM. DATASUS data is more accurate than police data due to the sys- tematic training of health providers in identifying homicide cases. The procedures to capture all external causes of death, including homicides, in the DATASUS data are more robust than the police data (Cano and Santos, 2001; Monteiro de Castro et. al., 2003). In addition, many injuries that result in deaths and can be typi�ed as homicides are not reported in o cial po- lice records but they are indeed reported in the DATASUS system. Although it is certain that both systems have undercounting of homicides, the DATASUS system has a broader range of categories to classify homicides. Because DATASUS is an integrated system that can dis- aggregate data at various levels there is better consistency in the de�nition and number of homicides across states and municipalities, in particular for more recent data. Since 1997 DATASUS reports appear to more closely follow police reports on crime, even though the DATASUS data appears to be consistently higher. ANNEX Choice of homicide rate as leading indicator of violent crime. While aware of the fact that violence is a complex phenomenon which manifests in various forms which may not be captured by the data: homicides rates represent the most extreme form of violence and by many is thought as “barometer� of all violent crime in society, with a relatively high level of comparability across time and by di erent levels of aggregation (Fajnzylber, Lederman and Loayza, 1999; Kovandzic and Vieraitis, 2006). In general, homicide data tends to be measured with relatively accuracy compared to other crime statistics, as it is di cult to underreport it. In addition, only homicides rates were available for the 1997-2008 for the 5,564 municipali- ties in Brazil. The data at the municipality level was collected from SIM/DATASUS with comparable homi- cide rates using the International Classi�cation of Disease (CID-10) from 1997 to 2008. Ad- ditional municipal level data was collected from IBGE and RAIS/CAGED. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME ANNEX II. KERNEL DENSITY DISTRIBUTIONS TESTS FOR EQUALITY Kolmogorov Smirnov tests for homicide rates 2008 v. 2003 Two-sample Kolmogorov-Smirnov test for equality of distribution functions Smaller group D P-value Corrected 0:00 0.0391 0.008 1:00 -0.3057 0 Combined K-S: 0.3057 0 0 We reject the null hypothesis of equality of distributions for the log homicide rate in 2008 v. 2003. Note: ties exist in combined dataset; there are 367 unique values out of 6296 observa- tions. Kolmogorov Smirnov tests for homicide rates 2008 v. 2003 in the Southeast region Smaller group D P-value Corrected 0:00 0.0103 0.91 1:00 -0.3253 0 Combined K-S: 0.3253 0 0 We reject the null hypothesis of equality of distributions for the log homicide rate in 2008 v. 2003. Note: ties exist in combined dataset; there are 232 unique values out of 1795 observa- tions. Kolmogorov Smirnov tests for homicide rates 2008 v. 2003 in the Northeast region Smaller group D P-value Corrected 0:00 0.1 0 1:00 -0.3067 0 Combined K-S: 0.3067 0 0 We reject the null hypothesis of equality of distributions for the log homicide rate in 2008 v. 2003. Note: ties exist in combined dataset; there are 95 unique values out of 422 observa- tions. ANNEX Kolmogorov Smirnov tests for homicide rates 2008 v. 2003 in the North region Smaller group D P-value Corrected 0:00 0.1095 0.08 1:00 -0.1689 0.002 Combined K-S: 0.1689 0.005 0.003 We reject the null hypothesis of equality of distributions for the log homicide rate in 2008 v. 2003. Note: ties exist in combined dataset; there are 95 unique values out of 422 observa- tions. Kolmogorov Smirnov tests for homicide rates 2008 v. 2003 in the South region Smaller group D P-value Corrected 0:00 0.0322 0.525 1:00 -0.3762 0 Combined K-S: 0.3762 0 0 We reject the null hypothesis of equality of distributions for the log homicide rate in 2008 v. 2003. Note: ties exist in combined dataset; there are 123 unique values out of 1244 observa- tions. Kolmogorov Smirnov tests for homicide rates 2008 v. 2003 in the Central West region Smaller group D P-value Corrected 0:00 0.0323 0.728 1:00 -0.2824 0 Combined K-S: 0.2824 0 0 We reject the null hypothesis of equality of distributions for the log homicide rate in 2008 v. 2003. Note: ties exist in combined dataset; there are 92 unique values out of 607 observa- tions. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME ANNEX III. FIXED EFFECTS REGRESSIONS OF THE LOG HOMICIDE RATE PER 100,000 FOR BRAZIL AND ITS REGIONS (1) (2) (3) (4) (5) (6) Variables Brazil Southeast Northeast North South Central West Log Males 15-19 0.397*** 0.485*** -0.440*** -0.696** 0.960*** 0.311** (0.0583) (0.102) (0.120) (0.311) (0.115) (0.144) Log Males 20-29 0.251*** 0.423*** 0.385*** 1.092*** -0.298** 0.100 (0.0583) (0.109) (0.119) (0.285) (0.116) (0.159) Log Population 0.160** 0.0437 0.835*** 0.224 0.224* 0.382** (0.0634) (0.123) (0.127) (0.277) (0.124) (0.151) Urban Population Share 10.52*** 22.75** 1.180 -18.20** 23.33** 6.862*** (1.254) (4.012) (2.516) (8.419) (11.81) (2.422) Population Density 0.000572*** 0.000517*** 0.000527*** 0.000455 0.000862*** 0.00197*** (1.82e-05) (2.35e-05) (5.66e-05) (0.000409) (8.64e-05) (0.000356) Population Density Squared -4.1e-08*** -3.71e-08*** -6.75e-08*** -6.44e-08 -1.49e-07*** -8.90e-07*** (1.87e-09) (2.25e-09) (9.80e-09) (1.82e-07) (2.96e-08) (2.08e-07) Log GDP 0.0808*** 0.0806*** 0.185*** 0.238*** 0.0314 -0.0102 (0.0118) (0.0192) (0.0242) (0.0636) (0.0254) (0.0324) Public High school dropout rate 0.344*** 0.159 0.373*** 0.388 0.0983 0.125 (0.0751) (0.147) (0.131) (0.375) (0.159) (0.182) Gin Coefficient 1.001** 2.625 0.870 3.692 0.387 -0.156 (0.459) (1.702) (0.660) (4.371) (1.414) (1.460) Formal Jobs Created 2.19e-07 -2.10e-07 8.24e-06*** 1.41e-05*** 3.92e-06*** 7.77e-06*** (3.04e-07) (3.21e-07) (2.24e-06) (5.37e-06) (1.34e-06) (2.18e-06) Formal Jobs Destroyed 5.40e-07* 4.57e-07 7.74e-06*** -2.59e-06 4.24e-06*** -1.79e-07 (3.04e-07) (3.19e-07) (2.13e-06) (4.57e-06) (1.50e-06) (2.10e-06) Formal Jobs Created males 15-29 -0.0491*** -0.104*** -0.0193 0.0204 -0.128*** -0.0615 (0.0131) (0.0364) (0.0162) (0.0599) (0.0396) (0.0445) Formal Jobs Destroyed males 15-29 0.0628*** 0.0240 0.0690*** 0.153*** 0.0352 0.0465 (0.0124) (0.0345) (0.0153) (0.0573) (0.0375) (0.0424) Constant -16.54*** -20.05*** -17.21*** -10.12*** -15.69*** -14.74*** (0.461) (1.646) (0.942) (3.123) (1.498) (1.392) Observations 19,121 6,802 5,318 874 4,304 1,823 R-squared 0.714 0.778 0.665 0.612 0.710 0.691 Source: Calderón and Chioda, 2011. Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Source: DATASUS, IBGE and RAIS, CAGED from 1997 to 2008 for all 5,564 municipali- ties. Information on school dropout rates is not available for 2006, and thus this year is dropped from the saturated model regressions. Year, region and state �xed e ects are included in the regressions. ANNEX ANNEX IV. OAXACA BLINDER DECOMPOSITION 2008 V. 2003 Brazil Southeast Northeast (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Variables overall endowments coefficients interaction overall endowments coefficients interaction overall endowments coefficients interaction Males 15-19 0.179*** -5.865*** -0.163*** 0.108 0.223 0.00550 -0.0198 -4.430 -0.184 (0.0374) (1.643) (0.0475) (0.0687) (3.308) (0.0817) (0.104) (3.875) (0.162) Males 20-29 0.0323** 11.70*** -0.0615** -0.0185 6.255 -0.0350 -0.00861 0.878 -0.00341 (0.0155) (1.985) (0.0268) (0.0228) (3.897) (0.0327) (0.0149) (4.185) (0.0165) Urban Population Share 0.00182* 0.0623** 0.000984 0.00840** 0.193** 0.00879* -0.00142 0.397** -0.00288 (0.000998) (0.0293) (0.000674) (0.00345) (0.0861) (0.00506) (0.00159) (0.158) (0.00300) GDP -0.0341*** 0.783** -0.0367** -0.0479** -0.583 0.0281 -0.0848*** 0.247 -0.0140 (0.0124) (0.313) (0.0147) (0.0215) (0.502) (0.0242) (0.0285) (0.645) (0.0366) Public School Dropout 0.0351*** -0.0735* -0.0123* 0.0588*** -0.0806 -0.0289 -0.00214 -0.163** 0.00272 rate (0.00553) (0.0378) (0.00640) (0.0182) (0.0628) (0.0226) (0.00266) (0.0778) (0.00338) Gini Coefficient 0.0778*** -0.734 -0.0463 0.0271 -21.71** -1.584** 0.115*** 1.924 0.0924 (0.0214) (0.546) (0.0345) (0.102) (10.77) (0.786) (0.0335) (1.351) (0.0650) Formal Jobs Added -0.00224* 0.0106** -0.00365** -0.00319* 0.00117 -0.000446 -0.00140 0.0435*** -0.0141** (0.00118) (0.00526) (0.00184) (0.00179) (0.00412) (0.00157) (0.00163) (0.0139) (0.00714) Formal Jobs Destroyed -0.000438 0.00536** -0.00204** 4395e-05 0.0124*** -0.00488*** -0.00314 -0.0131 0.00600 (0.00153) (0.00210) (0.000809) (0.00238) (0.00413) (0.00170) (0.00268) (0.0157) (0.00739) Formal Jobs Added Males 0.000639 0.0315 -0.00125 0.0306 0.353 -0.0196 -0.00876 -0.175 0.00454 15-29 (0.00750) (0.264) (0.0104) (0.0405) (0.955) (0.0531) (0.00424) (0.224) (0.00652) Formal Jobs Destroyed -0.0175** -0.321 0.0151 0.0125 -0.140 0.00923 -0.00644 -0.0861 0.00188 Males 15-29 (0.00866) (0.244) (0.0117) (0.0459) (0.899) (0.0591) (0.00631) (0.203) (0.00473) 2008 -6.203*** -6.081*** -6.383*** (0.0274) (0.0525) (0.0474) 2003 -6.228*** -6.210*** -6.287*** (0.0242) (0.0467) (0.0409) Difference 0.0248 0.128*** -0.0957** (0.0227) (0.0413) (0.0406) Endowments 0.309*** 0.176* 0.0770 (0.0480) (0.104) (0.111) Coefficients 0.0581 1.581** -0.127 (0.0402) (0.768) (0.118) Interaction -0.342*** -1.628** -0.0455 (0.0536) (0.765) (0.148) Constant 0.711 24.75** -3.605 (1.025) (12.01) (2.585) Observations 5,407 5,407 5,407 5,407 1,772 1,772 1,772 1,772 1,510 1,510 1,510 1,510 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Source: DATASUS, IBGE and RAIS, CAGED for 2003 and 2008 for all 5,564 municipalities. Information on school dropout rates is not available for 2006, and thus this year is dropped from the saturated model regressions. MAKING BRAZILIANS SAFER: ANALYZING THE DYNAMICS OF VIOLENT CRIME ANNEX V. CHANGES IN HOMICIDE RATES PER 100,000 INHABITANTS FROM 2003 TO 2008 BY STATE AND STATE CAPITAL Homicide Rate Homicide Rate Variables Capital 2003 2008 Change 2003 2008 Change Bahia 15.81 32.60 1.06 Salvador 27.74 59.10 1.13 Alagoas 34.92 59.79 0.71 Maceió 58.80 105.73 0.80 Parana 25.03 32.48 0.30 Curitiba 35.49 55.80 0.57 Maranhao 13.13 20.25 0.54 São Luís 29.61 43.20 0.45 Para 20.39 38.77 0.90 Belém 33.61 46.54 0.38 Rio Grande do Norte 13.92 23.08 0.66 Natal 22.32 30.76 0.38 Paraiba 17.62 27.42 0.56 João Pessoa 43.27 59.24 0.37 Amazonas 17.87 24.37 0.36 Manaus 28.13 37.73 0.34 Rio Grande do Sul 18.06 22.17 0.23 Porto Alegre 36.08 46.86 0.30 Ceara 19.56 23.78 0.22 Fortaleza 28.55 35.84 0.24 Goias 22.98 29.69 0.29 Goiânia 36.48 43.68 0.20 Espirito Santo 48.93 55.89 0.14 Vitória 71.40 73.40 0.03 Piaui 11.23 13.13 0.17 Teresina 27.60 27.04 (0.02) Rondonia 35.79 31.92 (0.11) Porto Velho 47.52 46.50 (0.02) Amapa 34.71 33.67 (0.03) Macapá 42.88 41.20 (0.04) Pernambuco 54.30 50.33 (0.07) Recife 90.26 84.65 (0.06) Distrito Federal 37.55 33.64 (0.10) Brasília 37.55 33.64 (0.10) Tocantins 18.43 18.47 0.00 Palmas 20.25 18.02 (0.11) Mato Grosso 33.81 31.43 (0.07) Cuiabá 48.22 42.50 (0.12) Santa Catarina 11.47 13.14 0.15 Florianópolis 26.10 22.79 (0.13) Sergipe 24.58 28.57 0.16 Aracaju 49.40 40.25 (0.19) Roraima 27.76 24.91 (0.10) Boa Vista 30.89 24.35 (0.21) Acre 21.68 19.55 (0.10) Rio Branco 36.35 28.44 (0.22) Mato Grosso do Sul 31.83 29.49 (0.07) Campo Grande 33.92 25.43 (0.25) Minas Gerais 20.54 16.68 (0.04) Belo Horizonte 56.67 41.55 (0.27) Rio de Janeiro 51.59 33.73 (0.35) Rio de Janeiro 55.36 30.95 (0.44) Sao Paulo 35.13 14.90 (0.58) São Paulo 51.64 14.89 (0.71) Source: DATASUS, IBGE and RAIS, CAGED for 2003 and 2008 for all 5,564 municipalities.